fix(lora): the import review takes a lease beside a restore

The LoRA import review reads the library, so like the maintenance and
restore previews it now runs under a request lease: it is refused with
409 busy while a restore runs, and a restore waits for it before it
switches the file. Tests pin that a restore is refused while an import
runs and that the review, the reviewed import and the admin page's scan
are refused while a restore runs.

.github/branch-protection.json

@@ -18,7 +18,8 @@
       "black and flake8",
       "frontend modules",
       "viewerjs",
-      "real ComfyUI compatibility"
+      "real ComfyUI compatibility",
+      "bandit (high severity)"
     ]
   },
   "enforce_admins": true,

.github/workflows/code-quality.yml

@@ -9,42 +9,15 @@ on:
   pull_request:
     branches: [main]
 
+# Superseded PR pushes cancel their older run. Every main push gets its own
+# group (run_id), so no main commit's run is ever cancelled or dropped as pending.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 jobs:
-  lint:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v7
-
-      - name: Set up Python 3.10
-        uses: actions/setup-python@v7
-        with:
-          python-version: "3.10"
-
-      - name: Cache pip dependencies
-        uses: actions/cache@v6
-        with:
-          path: ~/.cache/pip
-          key: ${{ runner.os }}-pip-quality-${{ hashFiles('requirements-dev.txt') }}
-          restore-keys: |
-            ${{ runner.os }}-pip-quality-
-
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install -r requirements-dev.txt
-
-      - name: Check formatting with Black
-        run: black --check --diff .
-
-      - name: Lint with flake8
-        run: |
-          # Fail on syntax errors and undefined names
-          flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics
-          # Warnings as non-blocking
-          flake8 . --count --exit-zero --max-line-length=88 --statistics --exclude=reference/,.git,__pycache__,web/
-
   security:
+    name: bandit (high severity)
     runs-on: ubuntu-latest
 
     steps:
@@ -58,5 +31,9 @@ jobs:
       - name: Install bandit
         run: pip install bandit[toml]
 
-      - name: Security scan
-        run: bandit -r . -ll --exclude=./tests,./reference,./venv -q || true
+      # Medium findings are reported without failing until they are triaged.
+      # High severity is a hard gate.
+      - name: Security report (medium and above, informational)
+        run: bandit -r . -ll -q --exclude ./tests,./reference,./venv,./tools/vendor/node_modules --exit-zero
+      - name: Security scan (high severity gate)
+        run: bandit -r . -lll -q --exclude ./tests,./reference,./venv,./tools/vendor/node_modules

.github/workflows/test.yml

@@ -8,6 +8,16 @@ on:
     branches: [main]
   push:
     branches: [main]
+  workflow_dispatch:
+
+# Superseded PR pushes cancel their older run. Every main push gets its own
+# group (run_id), so no main commit's run is ever cancelled or dropped as pending.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+env:
+  PYTHONFAULTHANDLER: "1"
 
 jobs:
   test:
@@ -56,6 +66,9 @@ jobs:
           node --input-type=module <<'JS' >> "$GITHUB_OUTPUT"
           import runtime from './tools/vendor/comfyui-runtime.json' with { type: 'json' };
           for (const [key, value] of Object.entries(runtime)) console.log(`${key}=${value}`);
+          import crossPack from './tools/vendor/cross-pack-runtime.json' with { type: 'json' };
+          console.log(`kikoRepository=${crossPack.repository}`);
+          console.log(`kikoRevision=${crossPack.revision}`);
           JS
       - name: Check out pinned ComfyUI
         uses: actions/checkout@v7
@@ -64,6 +77,13 @@ jobs:
           ref: ${{ steps.runtime.outputs.revision }}
           path: .ci-comfyui
           persist-credentials: false
+      - name: Check out pinned KikoTools for compatibility
+        uses: actions/checkout@v7
+        with:
+          repository: ${{ steps.runtime.outputs.kikoRepository }}
+          ref: ${{ steps.runtime.outputs.kikoRevision }}
+          path: .ci-kikotools
+          persist-credentials: false
       - uses: actions/setup-python@v7
         with:
           python-version: ${{ steps.runtime.outputs.pythonVersion }}
@@ -84,6 +104,17 @@ jobs:
           PM_COMFY_ROOT: ${{ github.workspace }}/.ci-comfyui
           PM_TEST_PYTHON: python
         run: npm run --prefix tools/vendor test:comfyui
+      - name: Real cached and headless execution history
+        env:
+          PM_COMFY_ROOT: ${{ github.workspace }}/.ci-comfyui
+          PM_TEST_PYTHON: python
+        run: npm run --prefix tools/vendor test:executions
+      - name: Independent and combined pack compatibility
+        env:
+          PM_COMFY_ROOT: ${{ github.workspace }}/.ci-comfyui
+          PM_KIKO_ROOT: ${{ github.workspace }}/.ci-kikotools
+          PM_TEST_PYTHON: python
+        run: npm run --prefix tools/vendor test:compat
       - name: Upload real-host reports and startup logs
         if: always()
         uses: actions/upload-artifact@v7
@@ -116,6 +147,32 @@ jobs:
           python -m coverage combine
           python -m coverage report --fail-under=85
 
+  perf-budget:
+    # The reference wall-clock measurement (TESTING.md "Performance budgets").
+    # Shared-runner timing varies, so it never gates a pull request: it runs on
+    # each push to main and on demand. A red run on main means fix the code.
+    name: perf budget
+    if: github.event_name == 'push' || github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+
+      - uses: actions/setup-python@v7
+        with:
+          python-version: "3.13"
+          cache: pip
+          cache-dependency-path: requirements-dev.txt
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements-dev.txt
+
+      - name: Workspace filter budget
+        env:
+          PROMPTMANAGER_PERF_BUDGET: "1"
+        run: python -m unittest -v tests.test_api_workspace_images.TestWorkspaceImages.test_filter_budget
+
   lint:
     name: black and flake8
     runs-on: ubuntu-latest
@@ -145,5 +202,8 @@ jobs:
             node --input-type=module --check < "$f" || exit 1
           done
 
+      - name: Install pinned shared test contracts
+        run: npm ci --prefix tools/vendor --ignore-scripts
+
       - name: Run JS tests
         run: node --test tests/js/*.test.js

.github/workflows/viewerjs-check.yml

@@ -10,38 +10,229 @@ on:
 permissions:
   contents: read
 
+# Superseded PR pushes cancel their older run. Every main push gets its own
+# group (run_id), so no main commit's run is ever cancelled or dropped as pending.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+# Timeouts are the measured job time plus at least 50%, and never under 20 minutes for a browser job,
+# because runner speed varies. Before sharding (main 2cecfcf, one worker): frontend coverage 32m10s,
+# page workflows firefox 18m59s, webkit 19m26s. First sharded run (PR #274 run 39712909047, 3 workers,
+# no npm cache, where npm ci stalled for 5m in six of eight jobs): coverage shards 11m19s/10m47s/10m49s
+# (tests 5m20s/4m42s/4m46s), merge 5m52s (report and check 4s), firefox shards 7m53s/9m27s (tests
+# 4m41s/3m29s), webkit shards 12m25s+ (cancelled at the old 12-minute limit after its 267 tests passed in
+# 6m29s) and 4m53s (tests 3m08s). With the npm cache hit (run 39725176320 at cbc8a61, npm ci 0-2s):
+# coverage shards 7m34s/6m04s/5m22s (tests 6m38s/4m53s/4m04s), merge 54s, firefox 7m57s/5m21s (tests
+# 7m08s/4m34s), webkit 8m16s/5m03s (tests 7m26s/3m58s); a cache miss adds up to 5m to npm ci, which
+# the 20-minute limits absorb.
 jobs:
-  viewerjs:
+  frontend-coverage:
+    name: frontend coverage (shard ${{ matrix.shard }}/3)
     runs-on: ubuntu-latest
-    timeout-minutes: 15
+    # Layout comparisons depend on the fonts each browser falls back to, so browser jobs run in the
+    # same pinned Playwright image as local runs (tag = @playwright/test in tools/vendor/package-lock.json;
+    # tests/vendor/browser-image.test.mjs checks it). Browsers and fonts come from the image.
+    container:
+      image: mcr.microsoft.com/playwright:v1.64.1-noble
+      options: --user 1001
+    timeout-minutes: 20
+    # Public-repo ubuntu-latest runners have 4 vCPUs: three Playwright workers, each with its own test
+    # server, leave one CPU for the servers and the runner.
+    env:
+      PM_TEST_WORKERS: 3
+    strategy:
+      fail-fast: false
+      matrix:
+        shard: [1, 2, 3]
     steps:
       - uses: actions/checkout@v7
-      - uses: actions/setup-python@v7
+      # The image's own Python 3.12 and Node 24: setup-python/setup-node install the runner's toolcache
+      # builds, whose entry points name host paths that do not exist in the container. The image has no
+      # ensurepip and the job has no root, so a checksum-pinned pip wheel bootstraps a user-space venv.
+      - name: Python environment
+        env:
+          PIP_WHEEL: https://files.pythonhosted.org/packages/f3/6e/1736e5b4ae2b778ef2f81c47d797de9f891d4d8acb047a24ca37a60294dd/pip-26.2.1-py3-none-any.whl
+          PIP_SHA256: 71138adf1f4ca900cdb7d289c21b7494329f2332b6d85f0e1c42108c0384ed3e
+        run: |
+          venv="$RUNNER_TEMP/venv" wheel="$RUNNER_TEMP/pip-26.2.1-py3-none-any.whl"
+          python3 -m venv --without-pip "$venv"
+          curl -fsSL --retry 3 -o "$wheel" "$PIP_WHEEL"
+          echo "$PIP_SHA256  $wheel" | sha256sum -c -
+          "$venv/bin/python" "$wheel/pip" install --no-index --no-cache-dir "$wheel"
+          "$venv/bin/python" -m pip install --no-cache-dir -r requirements-dev.txt
+          echo "$venv/bin" >> "$GITHUB_PATH"
+      - run: python --version && node --version && node --version | grep -q '^v24\.'
+      # npm's download cache, keyed on the lockfile; inside the container ~ is the job's HOME (/github/home),
+      # which npm also uses. node_modules itself is still built fresh by npm ci.
+      - name: Cache npm downloads
+        uses: actions/cache@v6
         with:
-          python-version: '3.12'
-          cache: pip
-          cache-dependency-path: requirements-dev.txt
-      - run: python -m pip install -r requirements-dev.txt
-      - uses: actions/setup-node@v7
-        with:
-          node-version: '24'
-          cache: npm
-          cache-dependency-path: tools/vendor/package-lock.json
+          path: ~/.npm
+          key: npm-${{ runner.os }}-playwright-v1.64.1-${{ hashFiles('tools/vendor/package-lock.json') }}
       - run: npm ci --prefix tools/vendor --ignore-scripts
-      - run: npm run --prefix tools/vendor check
-      - run: npm run --prefix tools/vendor test
-      - run: npx --prefix tools/vendor playwright install --with-deps chromium firefox webkit
-      - name: Frontend unit, viewer, workflow, and coverage checks
-        run: npm run --prefix tools/vendor test:coverage
-      - name: Firefox and WebKit page workflows
-        run: npm run --prefix tools/vendor test:workflows -- --project firefox --project webkit
-      - name: Upload browser traces and frontend coverage
+      # Shard 1 also runs the node units and the viewer suite, so the vendor pins and the tooling tests
+      # run on the lightest shard instead.
+      - if: matrix.shard == 3
+        run: npm run --prefix tools/vendor check
+      - if: matrix.shard == 3
+        run: npm run --prefix tools/vendor test
+      - name: Page workflows with coverage, shard ${{ matrix.shard }}/3
+        run: npm run --prefix tools/vendor test:coverage:shard -- ${{ matrix.shard }}/3
+      - name: Upload raw coverage
+        uses: actions/upload-artifact@v7
+        with:
+          name: coverage-raw-${{ matrix.shard }}
+          path: coverage/js/raw/
+          if-no-files-found: error
+          retention-days: 3
+      - name: Upload Chromium and touch traces
+        if: always()
+        uses: actions/upload-artifact@v7
+        with:
+          name: frontend-traces-shard-${{ matrix.shard }}
+          path: tools/vendor/test-results/
+          retention-days: 14
+
+  frontend-coverage-merge:
+    name: frontend coverage (merge)
+    needs: frontend-coverage
+    runs-on: ubuntu-latest
+    # Same image as the shards: the raw files name sources by the container's workspace path.
+    container:
+      image: mcr.microsoft.com/playwright:v1.64.1-noble
+      options: --user 1001
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v7
+      - run: node --version && node --version | grep -q '^v24\.'
+      # npm's download cache, keyed on the lockfile; inside the container ~ is the job's HOME (/github/home),
+      # which npm also uses. node_modules itself is still built fresh by npm ci.
+      - name: Cache npm downloads
+        uses: actions/cache@v6
+        with:
+          path: ~/.npm
+          key: npm-${{ runner.os }}-playwright-v1.64.1-${{ hashFiles('tools/vendor/package-lock.json') }}
+      - run: npm ci --prefix tools/vendor --ignore-scripts
+      # Each artifact lands in its own coverage-raw-N directory, so the next step can count them.
+      - uses: actions/download-artifact@v8
+        with:
+          pattern: coverage-raw-*
+          path: ${{ runner.temp }}/coverage-raw
+      # coverage:merge only checks that raw files exist; merging two of three shards would report
+      # partial coverage, so a missing or empty shard artifact fails here.
+      - name: Verify every shard's raw coverage arrived
+        shell: bash
+        env:
+          STAGED_RAW: ${{ runner.temp }}/coverage-raw
+          EXPECTED_SHARDS: 3
+        run: |
+          found=0
+          for shard in $(seq 1 "$EXPECTED_SHARDS"); do
+            directory="$STAGED_RAW/coverage-raw-$shard"
+            count=0
+            if [ -d "$directory" ]; then count=$(find "$directory" -maxdepth 1 -type f -name '*.json' | wc -l); fi
+            echo "coverage-raw-$shard: $count raw files"
+            if [ "$count" -gt 0 ]; then found=$((found + 1)); fi
+          done
+          echo "shard artifacts with raw coverage: $found of $EXPECTED_SHARDS"
+          if [ "$found" -ne "$EXPECTED_SHARDS" ]; then
+            echo "::error::Expected raw coverage from $EXPECTED_SHARDS shards, got $found of $EXPECTED_SHARDS"
+            exit 1
+          fi
+          staged=$(find "$STAGED_RAW" -mindepth 2 -maxdepth 2 -type f -name '*.json' | wc -l)
+          mkdir -p coverage/js/raw
+          # cp -n never overwrites; a clash shows up as fewer merged files than staged ones.
+          find "$STAGED_RAW" -mindepth 2 -maxdepth 2 -type f -name '*.json' -exec cp -n -t coverage/js/raw {} + || true
+          merged=$(find coverage/js/raw -maxdepth 1 -type f -name '*.json' | wc -l)
+          if [ "$merged" -ne "$staged" ]; then
+            echo "::error::Shards wrote clashing raw file names: $staged staged, $merged merged"
+            exit 1
+          fi
+          echo "merged $merged raw files"
+      - name: Frontend coverage report and baseline check
+        run: npm run --prefix tools/vendor coverage:merge
+      - name: Upload frontend coverage report
         if: always()
         uses: actions/upload-artifact@v7
         with:
           name: frontend-test-reports
           path: |
-            tools/vendor/test-results/
             coverage/js/
             !coverage/js/raw/**
           retention-days: 14
+
+  browser-workflows:
+    name: page workflows (${{ matrix.browser }}, shard ${{ matrix.shard }}/2)
+    runs-on: ubuntu-latest
+    # Layout comparisons depend on the fonts each browser falls back to, so browser jobs run in the
+    # same pinned Playwright image as local runs (tag = @playwright/test in tools/vendor/package-lock.json;
+    # tests/vendor/browser-image.test.mjs checks it). Browsers and fonts come from the image.
+    container:
+      image: mcr.microsoft.com/playwright:v1.64.1-noble
+      options: --user 1001
+    timeout-minutes: 20
+    # Public-repo ubuntu-latest runners have 4 vCPUs: three Playwright workers, each with its own test
+    # server, leave one CPU for the servers and the runner.
+    env:
+      PM_TEST_WORKERS: 3
+    strategy:
+      fail-fast: false
+      matrix:
+        browser: [firefox, webkit]
+        shard: [1, 2]
+    steps:
+      - uses: actions/checkout@v7
+      # The image's own Python 3.12 and Node 24: setup-python/setup-node install the runner's toolcache
+      # builds, whose entry points name host paths that do not exist in the container. The image has no
+      # ensurepip and the job has no root, so a checksum-pinned pip wheel bootstraps a user-space venv.
+      - name: Python environment
+        env:
+          PIP_WHEEL: https://files.pythonhosted.org/packages/f3/6e/1736e5b4ae2b778ef2f81c47d797de9f891d4d8acb047a24ca37a60294dd/pip-26.2.1-py3-none-any.whl
+          PIP_SHA256: 71138adf1f4ca900cdb7d289c21b7494329f2332b6d85f0e1c42108c0384ed3e
+        run: |
+          venv="$RUNNER_TEMP/venv" wheel="$RUNNER_TEMP/pip-26.2.1-py3-none-any.whl"
+          python3 -m venv --without-pip "$venv"
+          curl -fsSL --retry 3 -o "$wheel" "$PIP_WHEEL"
+          echo "$PIP_SHA256  $wheel" | sha256sum -c -
+          "$venv/bin/python" "$wheel/pip" install --no-index --no-cache-dir "$wheel"
+          "$venv/bin/python" -m pip install --no-cache-dir -r requirements-dev.txt
+          echo "$venv/bin" >> "$GITHUB_PATH"
+      - run: python --version && node --version && node --version | grep -q '^v24\.'
+      # npm's download cache, keyed on the lockfile; inside the container ~ is the job's HOME (/github/home),
+      # which npm also uses. node_modules itself is still built fresh by npm ci.
+      - name: Cache npm downloads
+        uses: actions/cache@v6
+        with:
+          path: ~/.npm
+          key: npm-${{ runner.os }}-playwright-v1.64.1-${{ hashFiles('tools/vendor/package-lock.json') }}
+      - run: npm ci --prefix tools/vendor --ignore-scripts
+      - name: Page workflows in ${{ matrix.browser }}, shard ${{ matrix.shard }}/2
+        run: npm run --prefix tools/vendor test:workflows -- --project ${{ matrix.browser }} --shard ${{ matrix.shard }}/2
+      - name: Upload browser traces
+        if: always()
+        uses: actions/upload-artifact@v7
+        with:
+          name: browser-traces-${{ matrix.browser }}-shard-${{ matrix.shard }}
+          path: tools/vendor/test-results/
+          retention-days: 14
+
+  # Branch protection requires the exact context "viewerjs". This job keeps that
+  # name and fails unless every frontend job above succeeded (every shard, the
+  # coverage merge and every browser shard), including when one was cancelled or skipped.
+  viewerjs:
+    needs: [frontend-coverage, frontend-coverage-merge, browser-workflows]
+    if: always()
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Require every frontend job to succeed
+        env:
+          COVERAGE_RESULT: ${{ needs.frontend-coverage.result }}
+          MERGE_RESULT: ${{ needs.frontend-coverage-merge.result }}
+          WORKFLOWS_RESULT: ${{ needs.browser-workflows.result }}
+        run: |
+          echo "frontend-coverage: $COVERAGE_RESULT"
+          echo "frontend-coverage-merge: $MERGE_RESULT"
+          echo "browser-workflows: $WORKFLOWS_RESULT"
+          test "$COVERAGE_RESULT" = success && test "$MERGE_RESULT" = success && test "$WORKFLOWS_RESULT" = success

.gitignore

@@ -211,3 +211,6 @@ coverage/
 
 # Disposable upstream checkout for real-host CI
 .ci-comfyui/
+
+# Old library files a Move library renamed (never deleted; #249 S6b)
+*.moved-to-*

README.md

@@ -162,6 +162,12 @@ when the fields are collapsed, so LoRA trigger outputs can connect directly.
 Encoding, storage, and CONDITIONING/text outputs match Classic. Both layouts
 support ComfyUI Classic and Nodes 2.0.
 
+**Prompt Manager (Split)** keeps the same editor and library visible together.
+Wide nodes place the library beside the editor; narrow nodes stack the panels.
+It shares Focused's category/tag autocomplete, search filters, preview-before-use,
+workflow persistence, and connectable prepend/append inputs. Encoding and storage
+match Classic and Focused.
+
 ### 🔬 ComfyUI Workflow Analysis
 
 - **PNG Metadata Parsing**: Advanced PNG chunk analysis to extract embedded ComfyUI data
@@ -525,7 +531,7 @@ Keep your database optimized with built-in maintenance tools:
 
 ### Database Location
 
-By default, the database is saved as `example_prompts.db` in the node directory. This file contains all your prompts and linked images and can be backed up or shared.
+By default, the database is saved as `prompts.db` in the extension directory. It contains your prompt library and image links. See the [database path compatibility note](#database-path-compatibility-211) before upgrading a deployment with a custom path.
 
 ### Searching Prompts
 
@@ -745,9 +751,51 @@ ComfyUI_PromptManager/
 
 ## Configuration
 
+### Where your library lives
+
+New installs keep the library (`prompts.db`, `config.json` with the CivitAI key, the
+database backups and the logs) in ComfyUI's system user folder
+`ComfyUI/user/__promptmanager/`. An update, a reinstall or ComfyUI-Manager replacing
+the node folder no longer touches it, and ComfyUI's `/userdata` web API does not
+serve `__` system folders. Thumbnails stay in `output/thumbnails`.
+
+- **Upgrading from an earlier version:** your library stays where it is, in the node
+  folder, until you move it; nothing is moved on its own. The workspace shows a notice
+  above every page and System › Overview a "Library location" row. **Move library**
+  copies the database with SQLite's backup API, verifies the copy (integrity check, every
+  workflow document, the same rows in every table), copies `config.json` and the backups,
+  switches to the new folder without a restart and only then renames the old files
+  `prompts.db.moved-to-__promptmanager` and `config.json.moved-to-__promptmanager` (never
+  deleted; the old backups stay too). If anything fails, the library stays in the node
+  folder, unchanged. **Keep it here** stops the notice; the Overview row stays. Take a
+  backup first (System › Backup & export or the dashboard's Backup) as with any change.
+- **Your own folder:** set `PROMPT_MANAGER_DATA_DIR` to an absolute folder outside
+  ComfyUI's output, input and temp folders and your image sources. If it is empty and a
+  node-folder library exists, that library stays in use until you move it there.
+- **An explicit `database.default_path`** in `config.json` and `PROMPT_MANAGER_CONFIG_PATH`
+  keep working as before (described below); such a library is not offered a move.
+
+### Thumbnails
+
+Previews in `output/thumbnails` are written as **WebP, quality 80, 512 px** by default
+(System › Appearance › Thumbnails sets the format, quality and size limit for every
+preview). Previews from older versions keep their old format (usually PNG) until you
+run **Convert outdated** (System › Overview or Appearance): it rewrites each one from
+its original and then removes the old file. Nothing converts on its own.
+
+**JPEG XL** (smaller still) can be chosen once the server has an encoder. Install one
+of these, then restart ComfyUI:
+
+- `pip install pillow-jxl-plugin` in ComfyUI's Python environment (simplest), or
+- an `ffmpeg` on PATH built with `libjxl`, or
+- `cjxl` from [libjxl](https://github.com/libjxl/libjxl) on PATH.
+
+Browsers that cannot show JPEG XL see the original images in the grid instead.
+
 ### Database Settings
 
-You can customize the database path by creating a `config.json` file in the extension root:
+You can customize the database path by creating a `config.json` file in the library's
+folder (the node folder for a library that was not moved):
 
 ```json
 {
@@ -767,6 +815,32 @@ Relative paths are resolved against the extension directory. Absolute paths are
 }
 ```
 
+### Database Path Compatibility (#211)
+
+PromptManager now honors the `database.default_path` setting in `config.json`
+when loaded as a ComfyUI custom node. Relative paths resolve against the extension
+directory, not ComfyUI's working directory; absolute paths are used as-is. Earlier
+versions could ignore this setting and use `<extension>/prompts.db` instead.
+Honoring a different configured path can therefore show an empty or different
+library. PromptManager does not automatically move or merge database files.
+
+Before upgrading or changing this setting:
+
+1. Identify the database your current installation actually uses and confirm its
+   library contents. Check both the configured path and the prior fallback
+   `<extension>/prompts.db`; keep track of which existing file you intend to use.
+2. Preserve a consistent backup of the active database with the dashboard's
+   Backup/Restore backup action or SQLite's online backup API. Do not copy only
+   the main file of a live SQLite database: committed data may still be in its
+   `-wal` file. For a filesystem copy, stop ComfyUI cleanly first and preserve any
+   remaining `-wal` file together with the main database.
+3. Set `database.default_path` to the existing intended database file. An absolute
+   path makes the selection explicit; do not move or overwrite either database
+   while choosing the path.
+4. Restart ComfyUI, then verify that the expected library appears and the active
+   database path resolves to the file you selected. Keep the backup until both
+   checks succeed.
+
 ### Gallery & Monitoring Settings
 
 Configure the automatic image detection and gallery system:
@@ -838,7 +912,7 @@ print(f"Average rating: {info['average_rating']}")
 # Create backup
 db.model.backup_database("backup_prompts.db")
 
-# The database file can be copied directly for backup
+# The online backup includes committed data still in the WAL.
 ```
 
 ## Development
@@ -938,6 +1012,15 @@ MIT License - see LICENSE file for details.
 
 ## Changelog
 
+### Unreleased (Library location)
+
+- **📁 Library outside the node folder**: new installs keep `prompts.db`, `config.json`, the database backups and the logs in `ComfyUI/user/__promptmanager/`, where an update or reinstall of the node cannot delete them and ComfyUI's `/userdata` API does not reach. An existing library stays in the node folder until you press **Move library** (a verified copy, then a live switch; the old files are renamed `*.moved-to-__promptmanager`, never deleted) or **Keep it here**. `PROMPT_MANAGER_DATA_DIR` names your own folder. After moving, restart ComfyUI once so the logs follow. Going back to an older version after a move: an older version does not look in `user/__promptmanager`, so stop ComfyUI and copy `user/__promptmanager/prompts.db` (with its `-wal` file, if one is left) and `config.json` into the node folder; the `*.moved-to-__promptmanager` files are the library as it was at the move, without later changes.
+
+### Unreleased (Workflows stored once)
+
+- **🗜️ Smaller database**: each distinct workflow and prompt graph is stored once, compressed, instead of again for every image. Images recorded by this version use it from the first start; the maintenance job "Store each workflow once" rewrites the existing records and then reclaims the space. A 20,000-image test library went from 1.4 GB to 60 MB. The job always takes a verified backup first: the first time, `prompts.db.pre-workflow-storage_<date time>` beside the database, kept until you delete it; on later runs, a normal pooled safety backup. No image or video file is opened or changed.
+- **⚠️ Downgrading**: images recorded by this version (and, after the job, all images) are stored in the new form. An older Prompt Manager version still opens and even restores such a database, but shows those images without workflows or prompt graphs until you update again. To go back to an older version, stop ComfyUI, delete `prompts.db-wal` and `prompts.db-shm`, and copy the **earliest** `prompts.db.pre-workflow-storage_<date time>` over `prompts.db`, or, if the job never ran, a backup taken before this update (records added since that backup are lost). Otherwise stay on this version.
+
 ### v3.3.1 (Proxy Routing Fix)
 
 - **🌐 Proxy support**: the dashboard, gallery, tag page and metadata viewer now work when ComfyUI is reached through a proxy mount such as SwarmUI's `/ComfyBackendDirect/`. Scripts, images, API calls, progress streams and gallery folder filters stay under the mount, and the canvas node's buttons use ComfyUI's `fileURL()`. Direct ComfyUI URLs are unchanged. Thanks to @ghosttigerllc-bit ([#164](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/164))

__init__.py

@@ -6,6 +6,7 @@ gallery system.
 This module provides prompt node types:
 - PromptManager: CLIP encoding node that outputs CONDITIONING
 - PromptManagerFocused: Editor-first layout using the same encoding contract
+- PromptManagerSplit: Side-by-side editor and library using the same contract
 - PromptManagerText: Text-only node that outputs STRING with prepend/append
   functionality
 
@@ -39,13 +40,20 @@ def get_version():
     return "unknown"
 
 
-from .prompt_manager import PromptManager, PromptManagerFocused  # noqa: E402
+from .prompt_manager import (  # noqa: E402
+    PromptManager,
+    PromptManagerFocused,
+    PromptManagerSplit,
+)
+from .prompt_manager_pair import PromptManagerPair  # noqa: E402
 from .prompt_manager_text import PromptManagerText  # noqa: E402
 from .prompt_search_list import PromptSearchList  # noqa: E402
 
 NODE_CLASS_MAPPINGS = {
     "PromptManager": PromptManager,
     "PromptManagerFocused": PromptManagerFocused,
+    "PromptManagerSplit": PromptManagerSplit,
+    "PromptManagerPair": PromptManagerPair,
     "PromptManagerText": PromptManagerText,
     "PromptSearchList": PromptSearchList,
 }
@@ -53,6 +61,8 @@ NODE_CLASS_MAPPINGS = {
 NODE_DISPLAY_NAME_MAPPINGS = {
     "PromptManager": "Prompt Manager (Classic)",
     "PromptManagerFocused": "Prompt Manager (Focused)",
+    "PromptManagerSplit": "Prompt Manager (Split)",
+    "PromptManagerPair": "Prompt Manager (+/-)",
     "PromptManagerText": "Prompt Manager Text",
     "PromptSearchList": "Prompt Search List",
 }
@@ -77,6 +87,22 @@ try:
     routes = config.routes
     api = PromptManagerAPI()
     api.add_routes(routes)
+    # Temp files of preview writes a crash interrupted (#317), off the
+    # import path: the walk may cover a large thumbnails folder.
+    import threading
+
+    threading.Thread(
+        target=api.sweep_thumbnail_temps,
+        name="PromptManagerPreviewTempSweep",
+        daemon=True,
+    ).start()
+    host_app = getattr(config.server_instance, "app", None)
+    if host_app is not None and hasattr(host_app, "on_cleanup"):
+        host_app.on_cleanup.append(api.cleanup_history_reads)
+        host_app.on_cleanup.append(api.cleanup_annotation_captures)
+    if host_app is not None and hasattr(host_app, "on_shutdown"):
+        # Open Needs review change streams end before the server waits on them.
+        host_app.on_shutdown.append(api.close_review_events)
 
 except Exception as e:
     # Log to internal logging system without console spam
@@ -108,6 +134,28 @@ try:
     _global_prompt_tracker = get_prompt_tracker(_global_db)
     _global_image_monitor = get_image_monitor(_global_db, _global_prompt_tracker)
 
+    # The Needs review pass for ambiguous new outputs runs through the API's
+    # capture store (the one the scan and the Tag dialog use): a linked output
+    # is never queued and its pending items leave the queue. It stops before
+    # that store is closed at shutdown.
+    _review_api = globals().get("api")
+    if _review_api is not None:
+        _global_image_monitor.set_review_recorder(_review_api.record_review_file)
+        # Every settled output's preview is queued for the API's background
+        # worker (#249): the workspace never asks to generate thumbnails.
+        _global_image_monitor.set_thumbnail_queue(_review_api.queue_capture_thumbnail)
+        _review_host = getattr(config.server_instance, "app", None)
+        if _review_host is not None and hasattr(_review_host, "on_cleanup"):
+
+            async def _stop_review_recording(
+                app, monitor=_global_image_monitor, api=_review_api
+            ):
+                monitor.set_review_recorder(None)
+                monitor.set_thumbnail_queue(None)
+                api.close_auto_thumbnails(final=True)
+
+            _review_host.on_cleanup.insert(0, _stop_review_recording)
+
     # Start monitoring
     _global_image_monitor.start_monitoring()
     _init_logger.info("Global image monitoring system started successfully")
@@ -148,6 +196,34 @@ except Exception as e:
             f"Failed to register usage hook: {e}"
         )
 
+
+def _history_db_factory() -> "PromptDatabase":
+    """Initialize history on its worker even if monitoring initialization failed."""
+    if _global_db is not None:
+        return _global_db
+    from .database.operations import PromptDatabase
+
+    return PromptDatabase()
+
+
+_history_registration = None
+try:
+    from .py.config import server_instance
+    from .utils.execution_history import register_history_hook
+
+    _history_registration = register_history_hook(server_instance, _history_db_factory)
+except Exception:
+    try:
+        from .utils.logging_config import get_logger
+
+        get_logger("prompt_manager.init").error("Failed to register execution history")
+    except Exception:
+        import logging
+
+        logging.getLogger("prompt_manager.init").error(
+            "Failed to register execution history"
+        )
+
 __all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"]
 
 # Print startup message with loaded tools

database/annotation_schema.py

@@ -0,0 +1,290 @@
+"""Image annotation schema group (#249): file locators, byte records, image tags.
+
+The group installs like ``image_favorites`` and ``saved_views``: optional in old
+backups, required once a database is current. It owns
+
+- ``media_annotation_metadata``: one row, the group's schema version and the
+  image index revision;
+- ``media_annotation_files``: one private canonical locator per physical file
+  (``normcase(join(realpath(root), relative))``, server-side only);
+- ``media_annotation_bytes``: one record per observed whole-file SHA-256 of a
+  file, with the size and ``mtime_ns`` verified when those bytes were last seen
+  and the tag set's revision;
+- ``media_image_tags``: the ordered image tag set of one byte record, joined
+  into the global ``tags`` registry so vocabulary maintenance carries image
+  links like prompt and pair links.
+
+Installation never backfills anything from prompt or pair tags and never
+touches the filesystem. Validation reads only: the catalog objects are compared
+after whitespace compaction, columns, foreign keys and indexes through the
+PRAGMAs, and persisted values in single bounded SQL passes.
+"""
+
+import re
+import sqlite3
+from typing import Optional
+
+from .image_favorites import MAX_KEY_LENGTH
+
+SCHEMA_VERSION = 1
+METADATA_TABLE = "media_annotation_metadata"
+FILES_TABLE = "media_annotation_files"
+BYTES_TABLE = "media_annotation_bytes"
+IMAGE_TAGS_TABLE = "media_image_tags"
+TABLES = (METADATA_TABLE, FILES_TABLE, BYTES_TABLE, IMAGE_TAGS_TABLE)
+MAX_LOCATOR_LENGTH = MAX_KEY_LENGTH
+_TIMESTAMP_BYTES = 64
+
+_TABLE_DDL = {
+    METADATA_TABLE: (
+        f"CREATE TABLE {METADATA_TABLE} (singleton_id INTEGER PRIMARY KEY "
+        "CHECK(singleton_id=1), schema_version INTEGER NOT NULL "
+        f"CHECK(schema_version={SCHEMA_VERSION}), "
+        "image_index_revision INTEGER NOT NULL DEFAULT 0 "
+        "CHECK(image_index_revision>=0))"
+    ),
+    FILES_TABLE: (
+        f"CREATE TABLE {FILES_TABLE} (id TEXT PRIMARY KEY CHECK(length(id)=36), "
+        "locator TEXT NOT NULL UNIQUE, created_at TEXT NOT NULL)"
+    ),
+    BYTES_TABLE: (
+        f"CREATE TABLE {BYTES_TABLE} (id TEXT PRIMARY KEY CHECK(length(id)=36), "
+        f"file_id TEXT NOT NULL REFERENCES {FILES_TABLE}(id) ON DELETE CASCADE, "
+        "sha256 TEXT NOT NULL CHECK(length(sha256)=64 AND sha256 NOT GLOB "
+        "'*[^0-9a-f]*'), size INTEGER NOT NULL CHECK(size>=0), "
+        "mtime_ns INTEGER NOT NULL, verified_at TEXT NOT NULL, "
+        "first_seen_at TEXT NOT NULL, tag_revision INTEGER NOT NULL DEFAULT 0 "
+        "CHECK(tag_revision>=0), UNIQUE(file_id, sha256))"
+    ),
+    IMAGE_TAGS_TABLE: (
+        f"CREATE TABLE {IMAGE_TAGS_TABLE} (byte_id TEXT NOT NULL REFERENCES "
+        f"{BYTES_TABLE}(id) ON DELETE CASCADE, tag_id INTEGER NOT NULL "
+        "REFERENCES tags(id) ON DELETE CASCADE, position INTEGER NOT NULL "
+        "CHECK(position>=0), PRIMARY KEY(byte_id, tag_id))"
+    ),
+}
+_INDEX_DDL = {
+    "idx_media_annotation_bytes_fingerprint": (
+        "CREATE INDEX idx_media_annotation_bytes_fingerprint ON "
+        f"{BYTES_TABLE}(file_id, size, mtime_ns)"
+    ),
+    "idx_media_image_tags_tag": (
+        f"CREATE INDEX idx_media_image_tags_tag ON {IMAGE_TAGS_TABLE}(tag_id)"
+    ),
+    "idx_media_image_tags_position": (
+        "CREATE UNIQUE INDEX idx_media_image_tags_position ON "
+        f"{IMAGE_TAGS_TABLE}(byte_id, position)"
+    ),
+}
+# SQLite names the indexes behind TEXT primary keys and UNIQUE constraints
+# ``sqlite_autoindex_<table>_<n>``; their catalog rows carry no SQL.
+_AUTOINDEXES = {FILES_TABLE: 2, BYTES_TABLE: 2, IMAGE_TAGS_TABLE: 1}
+# (name, declared type, not null, default, primary key position)
+_COLUMNS = {
+    METADATA_TABLE: (
+        ("singleton_id", "INTEGER", 0, None, 1),
+        ("schema_version", "INTEGER", 1, None, 0),
+        ("image_index_revision", "INTEGER", 1, "0", 0),
+    ),
+    FILES_TABLE: (
+        ("id", "TEXT", 0, None, 1),
+        ("locator", "TEXT", 1, None, 0),
+        ("created_at", "TEXT", 1, None, 0),
+    ),
+    BYTES_TABLE: (
+        ("id", "TEXT", 0, None, 1),
+        ("file_id", "TEXT", 1, None, 0),
+        ("sha256", "TEXT", 1, None, 0),
+        ("size", "INTEGER", 1, None, 0),
+        ("mtime_ns", "INTEGER", 1, None, 0),
+        ("verified_at", "TEXT", 1, None, 0),
+        ("first_seen_at", "TEXT", 1, None, 0),
+        ("tag_revision", "INTEGER", 1, "0", 0),
+    ),
+    IMAGE_TAGS_TABLE: (
+        ("byte_id", "TEXT", 1, None, 1),
+        ("tag_id", "INTEGER", 1, None, 2),
+        ("position", "INTEGER", 1, None, 0),
+    ),
+}
+# (column, parent table, parent column, on delete)
+_FOREIGN_KEYS = {
+    METADATA_TABLE: set(),
+    FILES_TABLE: set(),
+    BYTES_TABLE: {("file_id", FILES_TABLE, "id", "CASCADE")},
+    IMAGE_TAGS_TABLE: {
+        ("byte_id", BYTES_TABLE, "id", "CASCADE"),
+        ("tag_id", "tags", "id", "CASCADE"),
+    },
+}
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql).lower()
+
+
+def _uuid(column: str) -> str:
+    """SQL: *column* is a lowercase hyphenated 36-character UUID string."""
+    return (
+        f"typeof({column}) = 'text' AND length({column}) = 36 "
+        f"AND substr({column}, 9, 1) = '-' AND substr({column}, 14, 1) = '-' "
+        f"AND substr({column}, 19, 1) = '-' AND substr({column}, 24, 1) = '-' "
+        f"AND length(replace({column}, '-', '')) = 32 "
+        f"AND replace({column}, '-', '') NOT GLOB '*[^0-9a-f]*'"
+    )
+
+
+def _timestamp(column: str) -> str:
+    return (
+        f"typeof({column}) = 'text' "
+        f"AND length(CAST({column} AS BLOB)) <= {_TIMESTAMP_BYTES}"
+    )
+
+
+def _non_negative(column: str) -> str:
+    return f"typeof({column}) = 'integer' AND {column} >= 0"
+
+
+# Every row of a table must satisfy its predicate; one pass each, LIMIT 1.
+_ROW_PREDICATES = {
+    FILES_TABLE: " AND ".join(
+        (
+            _uuid("id"),
+            "typeof(locator) = 'text'",
+            # UTF-8 is at most 4 bytes per character; the decoded length of
+            # the few candidates above the byte bound is checked in Python.
+            f"length(CAST(locator AS BLOB)) <= {MAX_LOCATOR_LENGTH * 4}",
+            _timestamp("created_at"),
+        )
+    ),
+    BYTES_TABLE: " AND ".join(
+        (
+            _uuid("id"),
+            _uuid("file_id"),
+            "typeof(sha256) = 'text' AND length(sha256) = 64 "
+            "AND sha256 NOT GLOB '*[^0-9a-f]*'",
+            _non_negative("size"),
+            "typeof(mtime_ns) = 'integer'",
+            _timestamp("verified_at"),
+            _timestamp("first_seen_at"),
+            _non_negative("tag_revision"),
+        )
+    ),
+    IMAGE_TAGS_TABLE: " AND ".join(
+        (_uuid("byte_id"), "typeof(tag_id) = 'integer'", _non_negative("position"))
+    ),
+}
+
+
+def _expected_catalog() -> set:
+    objects = {("table", name, _compact(sql)) for name, sql in _TABLE_DDL.items()}
+    objects |= {("index", name, _compact(sql)) for name, sql in _INDEX_DDL.items()}
+    objects |= {
+        ("index", f"sqlite_autoindex_{table}_{n}", "")
+        for table, count in _AUTOINDEXES.items()
+        for n in range(1, count + 1)
+    }
+    return objects
+
+
+def _catalog(conn: sqlite3.Connection) -> list:
+    names = TABLES + tuple(_INDEX_DDL)
+    marks = ",".join("?" * len(names))
+    table_marks = ",".join("?" * len(TABLES))
+    return conn.execute(  # only "?" placeholders are interpolated
+        "SELECT type,name,sql FROM sqlite_master "
+        f"WHERE name IN ({marks}) OR tbl_name IN ({table_marks})",  # nosec B608
+        names + TABLES,
+    ).fetchall()
+
+
+def _structure_error(conn: sqlite3.Connection) -> Optional[str]:
+    for table, expected in _COLUMNS.items():
+        columns = tuple(
+            (row[1], row[2].upper(), row[3], row[4], row[5])
+            for row in conn.execute(f"PRAGMA table_info({table})")
+        )
+        if columns != expected:
+            return f"Image annotations table {table} has invalid columns"
+        foreign_keys = {
+            (row[3], row[2], row[4], row[6])
+            for row in conn.execute(f"PRAGMA foreign_key_list({table})")
+        }
+        if foreign_keys != _FOREIGN_KEYS[table]:
+            return f"Image annotations table {table} has invalid foreign keys"
+        if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
+            return f"Image annotations table {table} contains foreign key violations"
+    return None
+
+
+def _value_error(conn: sqlite3.Connection) -> Optional[str]:
+    # Only this module's constants are interpolated into the SQL below.
+    rows = conn.execute(
+        "SELECT COUNT(*), MIN(typeof(singleton_id) = 'integer' "  # nosec B608
+        "AND singleton_id = 1 AND typeof(schema_version) = 'integer' "
+        f"AND schema_version = {SCHEMA_VERSION} "
+        f"AND {_non_negative('image_index_revision')}) FROM {METADATA_TABLE}"
+    ).fetchone()
+    if rows[0] != 1 or rows[1] != 1:
+        return "Image annotations metadata is invalid"
+    for table, predicate in _ROW_PREDICATES.items():
+        if conn.execute(
+            f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1"  # nosec B608
+        ).fetchone():
+            return f"Image annotations table {table} contains invalid values"
+    # length() stops at an embedded NUL, so the SQL pass bounds bytes and the
+    # few candidates over the character bound are measured decoded here.
+    for (raw,) in conn.execute(
+        f"SELECT CAST(locator AS BLOB) FROM {FILES_TABLE} "  # nosec B608
+        f"WHERE length(CAST(locator AS BLOB)) > {MAX_LOCATOR_LENGTH}"
+    ):
+        if len(bytes(raw).decode("utf-8", "replace")) > MAX_LOCATOR_LENGTH:
+            return f"Image annotations table {FILES_TABLE} contains invalid values"
+    return None
+
+
+def annotation_schema_error(
+    conn: sqlite3.Connection, *, allow_missing: bool = True
+) -> Optional[str]:
+    """Reject partial or unsupported annotation schemas and invalid rows.
+
+    Reads only: nothing is ever repaired. A database without any of the
+    group's objects is acceptable for backups that predate it unless
+    ``allow_missing`` is false.
+    """
+    objects = _catalog(conn)
+    if not objects:
+        return None if allow_missing else "Missing image annotation schema"
+    present = {name for kind, name, _sql in objects if kind == "table"}
+    if present != set(TABLES):
+        return "Incomplete image annotation schema"
+    actual = {(kind, name, _compact(sql or "")) for kind, name, sql in objects}
+    if actual != _expected_catalog():
+        return "Image annotations has an invalid schema definition"
+    return _structure_error(conn) or _value_error(conn)
+
+
+def migrate_annotation_state(conn: sqlite3.Connection) -> None:
+    """Install the group atomically; validate before and after creating it."""
+    conn.execute("SAVEPOINT media_annotation_schema")
+    try:
+        error = annotation_schema_error(conn)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        for sql in _TABLE_DDL.values():
+            conn.execute(sql.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        for sql in _INDEX_DDL.values():
+            conn.execute(sql.replace("INDEX", "INDEX IF NOT EXISTS", 1))
+        conn.execute(
+            f"INSERT OR IGNORE INTO {METADATA_TABLE} "
+            "(singleton_id, schema_version, image_index_revision) VALUES (1, ?, 0)",
+            (SCHEMA_VERSION,),
+        )
+        error = annotation_schema_error(conn, allow_missing=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute("RELEASE media_annotation_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO media_annotation_schema")
+        conn.execute("RELEASE media_annotation_schema")
+        raise

database/annotations.py

@@ -0,0 +1,461 @@
+"""Per-image tag sets (#249): byte-exact storage in the shared vocabulary.
+
+A tag set belongs to one byte record (``media_annotation_bytes``): the
+whole-file SHA-256 of one physical file, owned by the file's private locator
+(``media_annotation_files``). Callers pass verified facts (digest, size,
+``mtime_ns``); nothing here reads the filesystem or resolves an owner through
+prompt links, so tagging one image never changes another image, a prompt or a
+pair.
+
+Commands run in one ``_pair_transaction`` fenced by the storage epoch, like the
+native Library and vocabulary commands. A save is a full replacement guarded
+by the set's ``tag_revision``; the identical set sent again (a lost-response
+retry, or the current set at its own revision) answers ``replayed: true`` and
+writes nothing. ``add_image_tags`` unions into the existing set, keeping the
+stored spelling. Names enter the global ``tags`` registry exactly as spelled,
+so vocabulary rename, merge and delete carry image links with prompt and pair
+links (``PromptDatabase._move_tag_links``).
+
+Cheap reads (spec amendment S1) treat a byte record as *current* only while
+the file's present ``(size, mtime_ns)`` equals the record's verified
+fingerprint; a file without a current record shows no image tags.
+"""
+
+import re
+import sqlite3
+from contextlib import contextmanager
+from typing import Iterator
+from uuid import uuid4
+
+from .annotation_schema import (
+    BYTES_TABLE,
+    FILES_TABLE,
+    IMAGE_TAGS_TABLE,
+    MAX_LOCATOR_LENGTH,
+    METADATA_TABLE,
+)
+from .library_commands import _UUID, LibraryCommandError, _validate_epoch
+from .models import utc_now_iso
+from .vocabulary import validate_vocabulary_tag_name
+
+MAX_IMAGE_TAGS = 100
+_LOOKUP_CHUNK = 500
+_DIGEST = re.compile(r"[0-9a-f]{64}")
+_MAX_INT = 2**63 - 1
+_MIN_INT = -(2**63)
+# Only this module's table constants are interpolated into the SQL below.
+_OWNER = f"SELECT id FROM {FILES_TABLE} WHERE locator=?"  # nosec B608
+_INSERT_OWNER = (
+    f"INSERT INTO {FILES_TABLE}(id,locator,created_at) VALUES(?,?,?)"  # nosec B608
+)
+_BYTE = (
+    f"SELECT id,tag_revision,size,mtime_ns FROM {BYTES_TABLE} "  # nosec B608
+    "WHERE file_id=? AND sha256=?"
+)
+_INSERT_BYTE = (
+    f"INSERT INTO {BYTES_TABLE}(id,file_id,sha256,size,mtime_ns,"  # nosec B608
+    "verified_at,first_seen_at,tag_revision) VALUES(?,?,?,?,?,?,?,0)"
+)
+_REFRESH_BYTE = (
+    f"UPDATE {BYTES_TABLE} SET size=?,mtime_ns=?,verified_at=? WHERE id=?"  # nosec B608
+)
+_REFRESH_BY_LOCATOR = (
+    f"UPDATE {BYTES_TABLE} SET size=?,mtime_ns=?,verified_at=? "  # nosec B608
+    f"WHERE sha256=? AND file_id=({_OWNER})"
+)
+_BYTE_TAGS = (
+    f"SELECT t.name FROM {IMAGE_TAGS_TABLE} mit "  # nosec B608
+    "JOIN tags t ON t.id=mit.tag_id WHERE mit.byte_id=? ORDER BY mit.position"
+)
+_CLEAR_TAGS = f"DELETE FROM {IMAGE_TAGS_TABLE} WHERE byte_id=?"  # nosec B608
+_INSERT_TAG = (
+    f"INSERT INTO {IMAGE_TAGS_TABLE}(byte_id,tag_id,position) "  # nosec B608
+    "VALUES(?,?,?)"
+)
+_BUMP_REVISION = (
+    f"UPDATE {BYTES_TABLE} SET tag_revision=tag_revision+1 WHERE id=?"  # nosec B608
+)
+_REVISION = f"SELECT tag_revision FROM {BYTES_TABLE} WHERE id=?"  # nosec B608
+_BUMP_INDEX = (
+    f"UPDATE {METADATA_TABLE} SET "  # nosec B608
+    "image_index_revision=image_index_revision+1 WHERE singleton_id=1"
+)
+_READ = (
+    "SELECT b.id,b.tag_revision,"  # nosec B608
+    f"(SELECT COUNT(*) FROM {BYTES_TABLE} o WHERE o.file_id=f.id AND o.sha256!=?) "
+    f"FROM {FILES_TABLE} f LEFT JOIN {BYTES_TABLE} b "
+    "ON b.file_id=f.id AND b.sha256=? WHERE f.locator=?"
+)
+_FINGERPRINTS = (
+    "SELECT f.locator,b.id,b.size,b.mtime_ns,b.verified_at "  # nosec B608
+    f"FROM {FILES_TABLE} f JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    "WHERE f.locator IN ({marks})"
+)
+_TAGS_OF = (
+    f"SELECT mit.byte_id,t.name FROM {IMAGE_TAGS_TABLE} mit "  # nosec B608
+    "JOIN tags t ON t.id=mit.tag_id WHERE mit.byte_id IN ({marks}) "
+    "ORDER BY mit.byte_id,mit.position"
+)
+_RECORDS = (
+    "SELECT f.locator,b.id,b.size,b.mtime_ns,t.name "  # nosec B608
+    f"FROM {FILES_TABLE} f JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"LEFT JOIN {IMAGE_TAGS_TABLE} mit ON mit.byte_id=b.id "
+    "LEFT JOIN tags t ON t.id=mit.tag_id "
+    "ORDER BY f.locator,b.first_seen_at,b.id,mit.position"
+)
+
+
+class AnnotationCommandError(LibraryCommandError):
+    """An image tag command lost a race or hit the set's bound.
+
+    ``code`` is ``tag-conflict`` (the set changed since it was read) or
+    ``tag-limit`` (the union would exceed ``MAX_IMAGE_TAGS``); ``current`` is
+    the stored set ``{sha256, tags, tag_revision}`` read inside the
+    transaction, so the caller can show it.
+    """
+
+    def __init__(self, code: str, epoch: str, current: dict | None = None):
+        super().__init__(code, epoch)
+        self.current = current
+
+
+class ImageTagLimitError(ValueError):
+    """More than ``MAX_IMAGE_TAGS`` distinct names in one set."""
+
+
+def normalize_image_tags(names: object) -> list[str]:
+    """Trim and validate each name, fold case-insensitive repeats, cap at 100.
+
+    The first occurrence keeps its spelling and position. Raises ``ValueError``
+    for anything but a list of valid vocabulary names, and its subclass
+    ``ImageTagLimitError`` for more than ``MAX_IMAGE_TAGS`` distinct names.
+    """
+    if not isinstance(names, list):
+        raise ValueError("tags must be a list of strings")
+    result: list[str] = []
+    seen: set[str] = set()
+    for value in names:
+        name = validate_vocabulary_tag_name(value)
+        key = name.casefold()
+        if key in seen:
+            continue
+        seen.add(key)
+        result.append(name)
+    if len(result) > MAX_IMAGE_TAGS:
+        raise ImageTagLimitError(f"An image can carry at most {MAX_IMAGE_TAGS} tags")
+    return result
+
+
+def _validate_locator(locator: object) -> None:
+    if not isinstance(locator, str) or not 0 < len(locator) <= MAX_LOCATOR_LENGTH:
+        raise ValueError(
+            f"locator must be a nonblank path of at most {MAX_LOCATOR_LENGTH} chars"
+        )
+
+
+def _validate_digest(sha256: object) -> None:
+    if not isinstance(sha256, str) or not _DIGEST.fullmatch(sha256):
+        raise ValueError("sha256 must be 64 lowercase hexadecimal characters")
+
+
+def _validate_int(value: object, name: str, minimum: int) -> None:
+    if type(value) is not int or not minimum <= value <= _MAX_INT:
+        raise ValueError(f"{name} must be an integer not below {minimum}")
+
+
+def _validate_target(locator: object, sha256: object, size: object, mtime_ns: object):
+    _validate_locator(locator)
+    _validate_digest(sha256)
+    _validate_int(size, "size", 0)
+    _validate_int(mtime_ns, "mtime_ns", _MIN_INT)
+
+
+def _validate_fingerprint(locator: object, fingerprint: object) -> None:
+    _validate_locator(locator)
+    if (
+        not isinstance(fingerprint, tuple)
+        or len(fingerprint) != 2
+        or any(type(value) is not int for value in fingerprint)
+    ):
+        raise ValueError("stats must map locators to (size, mtime_ns) integer pairs")
+
+
+def _new_id() -> str:
+    """A canonical UUID; the schema's CHECK lets NULL through, this does not."""
+    value = str(uuid4())
+    if not _UUID.fullmatch(value):
+        raise ValueError("Annotation ids must be canonical UUIDs")
+    return value
+
+
+def _receipt(epoch: str, sha256: str, tags: list, revision: int, replayed: bool):
+    return {
+        "library_epoch": epoch,
+        "sha256": sha256,
+        "tags": list(tags),
+        "tag_revision": revision,
+        "replayed": replayed,
+    }
+
+
+class AnnotationOperations:
+    """PromptDatabase mixin: image tag sets, fingerprints and cheap reads.
+
+    ``_`` helpers take the open ``_pair_transaction`` connection.
+    """
+
+    @contextmanager
+    def _annotation_read(self) -> Iterator[sqlite3.Connection]:
+        """One deferred read transaction, so paired SELECTs see one snapshot."""
+        conn = self.model.get_connection()
+        if conn.in_transaction:
+            raise sqlite3.ProgrammingError(
+                "Annotation reads require their own snapshot"
+            )
+        with conn:
+            conn.execute("BEGIN")
+            yield conn
+
+    @staticmethod
+    def _annotation_owner(conn: sqlite3.Connection, locator: str, now: str) -> str:
+        row = conn.execute(_OWNER, (locator,)).fetchone()
+        if row is not None:
+            return row[0]
+        file_id = _new_id()
+        conn.execute(_INSERT_OWNER, (file_id, locator, now))
+        return file_id
+
+    @staticmethod
+    def _annotation_byte(
+        conn: sqlite3.Connection,
+        file_id: str,
+        sha256: str,
+        size: int,
+        mtime_ns: int,
+        now: str,
+    ) -> tuple[str, int, bool, bool]:
+        """The byte record for these bytes, refreshed or created.
+
+        Returns ``(id, tag_revision, created, refingerprinted)``; the last is
+        True when an existing record's verified ``(size, mtime_ns)`` changed,
+        which moves its currency for cheap reads (the caller bumps the index
+        revision when nothing else does).
+        """
+        row = conn.execute(_BYTE, (file_id, sha256)).fetchone()
+        if row is None:
+            byte_id = _new_id()
+            conn.execute(
+                _INSERT_BYTE, (byte_id, file_id, sha256, size, mtime_ns, now, now)
+            )
+            return byte_id, 0, True, False
+        conn.execute(_REFRESH_BYTE, (size, mtime_ns, now, row[0]))
+        return row[0], row[1], False, (row[2], row[3]) != (size, mtime_ns)
+
+    @staticmethod
+    def _byte_tags(conn: sqlite3.Connection, byte_id: str) -> list[str]:
+        return [row[0] for row in conn.execute(_BYTE_TAGS, (byte_id,))]
+
+    def _write_image_tags(
+        self, conn: sqlite3.Connection, byte_id: str, tags: list[str]
+    ) -> int:
+        """Rewrite the junction in order, bump both revisions; return the new one."""
+        tag_ids = self._ensure_tags(conn, tags)
+        conn.execute(_CLEAR_TAGS, (byte_id,))
+        conn.executemany(
+            _INSERT_TAG,
+            [(byte_id, tag_ids[name], position) for position, name in enumerate(tags)],
+        )
+        conn.execute(_BUMP_REVISION, (byte_id,))
+        conn.execute(_BUMP_INDEX)
+        return conn.execute(_REVISION, (byte_id,)).fetchone()[0]
+
+    def _current_set(self, conn, sha256, byte_id, revision) -> dict:
+        return {
+            "sha256": sha256,
+            "tags": self._byte_tags(conn, byte_id),
+            "tag_revision": revision,
+        }
+
+    def read_image_tags(self, locator: str, sha256: str) -> dict:
+        """The stored set of these exact bytes, or an empty set at revision 0.
+
+        ``prior_count`` is the number of other byte records at the locator
+        (earlier or later bytes of the same file).
+        """
+        _validate_locator(locator)
+        _validate_digest(sha256)
+        with self._annotation_read() as conn:
+            row = conn.execute(_READ, (sha256, sha256, locator)).fetchone()
+            if row is None:
+                return {"tags": [], "tag_revision": 0, "prior_count": 0}
+            tags = self._byte_tags(conn, row[0]) if row[0] is not None else []
+            return {"tags": tags, "tag_revision": row[1] or 0, "prior_count": row[2]}
+
+    def replace_image_tags(
+        self,
+        *,
+        locator: str,
+        sha256: str,
+        size: int,
+        mtime_ns: int,
+        expected_epoch: str,
+        expected_tag_revision: int,
+        tags: list,
+    ) -> dict:
+        """Replace the set of these exact bytes, guarded by epoch and revision.
+
+        Creates the owner and byte record when the bytes are new (an explicit
+        empty set is a real save). The identical set at the current revision,
+        or one behind it (a lost-response retry), is ``replayed`` and writes
+        nothing; any other revision mismatch is ``tag-conflict`` with the
+        current set and no writes.
+        """
+        _validate_target(locator, sha256, size, mtime_ns)
+        _validate_epoch(expected_epoch)
+        _validate_int(expected_tag_revision, "expected_tag_revision", 0)
+        tags = normalize_image_tags(tags)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            now = utc_now_iso()
+            file_id = self._annotation_owner(conn, locator, now)
+            byte_id, revision, created, refingerprinted = self._annotation_byte(
+                conn, file_id, sha256, size, mtime_ns, now
+            )
+            current = self._byte_tags(conn, byte_id)
+            if (
+                not created
+                and current == tags
+                and expected_tag_revision in (revision, revision - 1)
+            ):
+                if refingerprinted:
+                    conn.execute(_BUMP_INDEX)
+                return _receipt(epoch, sha256, current, revision, True)
+            if expected_tag_revision != revision:
+                raise AnnotationCommandError(
+                    "tag-conflict",
+                    epoch,
+                    {"sha256": sha256, "tags": current, "tag_revision": revision},
+                )
+            revision = self._write_image_tags(conn, byte_id, tags)
+            return _receipt(epoch, sha256, tags, revision, False)
+
+    def add_image_tags(
+        self,
+        *,
+        locator: str,
+        sha256: str,
+        size: int,
+        mtime_ns: int,
+        expected_epoch: str,
+        tags: list,
+    ) -> dict:
+        """Union ``tags`` into the set of these exact bytes, keeping stored spelling.
+
+        ``added`` counts the names that were new (case-insensitively). A union
+        over ``MAX_IMAGE_TAGS`` is ``tag-limit`` with the current set and no
+        writes; a union that adds nothing to an existing record is ``replayed``.
+        """
+        _validate_target(locator, sha256, size, mtime_ns)
+        _validate_epoch(expected_epoch)
+        tags = normalize_image_tags(tags)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            now = utc_now_iso()
+            file_id = self._annotation_owner(conn, locator, now)
+            byte_id, revision, created, refingerprinted = self._annotation_byte(
+                conn, file_id, sha256, size, mtime_ns, now
+            )
+            current = self._byte_tags(conn, byte_id)
+            present = {name.casefold() for name in current}
+            added = [name for name in tags if name.casefold() not in present]
+            merged = current + added
+            if len(merged) > MAX_IMAGE_TAGS:
+                raise AnnotationCommandError(
+                    "tag-limit",
+                    epoch,
+                    {"sha256": sha256, "tags": current, "tag_revision": revision},
+                )
+            if not created and not added:
+                if refingerprinted:
+                    conn.execute(_BUMP_INDEX)
+                return {**_receipt(epoch, sha256, current, revision, True), "added": 0}
+            revision = self._write_image_tags(conn, byte_id, merged)
+            return {
+                **_receipt(epoch, sha256, merged, revision, False),
+                "added": len(added),
+            }
+
+    def refresh_image_fingerprint(
+        self, locator: str, sha256: str, size: int, mtime_ns: int
+    ) -> bool:
+        """Record that these exact bytes were just seen with this stat.
+
+        True when a byte record was refreshed; nothing is created and no tag
+        or tag revision changes. The image index revision moves because the
+        record's currency for cheap reads may have changed.
+        """
+        _validate_target(locator, sha256, size, mtime_ns)
+        with self.model.get_connection() as conn:
+            cursor = conn.execute(
+                _REFRESH_BY_LOCATOR, (size, mtime_ns, utc_now_iso(), sha256, locator)
+            )
+            if cursor.rowcount > 0:
+                conn.execute(_BUMP_INDEX)
+            return cursor.rowcount > 0
+
+    def current_image_tags(self, stats: dict) -> dict:
+        """Image tags of the files whose current ``(size, mtime_ns)`` is verified.
+
+        ``stats`` maps locators to their present fingerprint. A locator is
+        answered (possibly with an empty set) only when a byte record carries
+        exactly that verified fingerprint; the most recently verified record
+        wins when several do. Lookups go in chunks of 500 locators.
+        """
+        if not isinstance(stats, dict):
+            raise ValueError(
+                "stats must map locators to (size, mtime_ns) integer pairs"
+            )
+        for locator, fingerprint in stats.items():
+            _validate_fingerprint(locator, fingerprint)
+        locators = list(stats)
+        chosen: dict[str, tuple[tuple[str, str], str]] = {}
+        result: dict[str, list[str]] = {}
+        with self._annotation_read() as conn:
+            for start in range(0, len(locators), _LOOKUP_CHUNK):
+                chunk = locators[start : start + _LOOKUP_CHUNK]
+                marks = ",".join("?" * len(chunk))
+                for locator, byte_id, size, mtime_ns, verified_at in conn.execute(
+                    _FINGERPRINTS.format(marks=marks), chunk
+                ):
+                    if (size, mtime_ns) != stats[locator]:
+                        continue
+                    key = (verified_at, byte_id)
+                    if locator not in chosen or key > chosen[locator][0]:
+                        chosen[locator] = (key, byte_id)
+            owners = {byte_id: locator for locator, (_key, byte_id) in chosen.items()}
+            result = {locator: [] for locator in chosen}
+            byte_ids = list(owners)
+            for start in range(0, len(byte_ids), _LOOKUP_CHUNK):
+                chunk = byte_ids[start : start + _LOOKUP_CHUNK]
+                marks = ",".join("?" * len(chunk))
+                for byte_id, name in conn.execute(_TAGS_OF.format(marks=marks), chunk):
+                    result[owners[byte_id]].append(name)
+        return result
+
+    def iter_image_tag_records(self) -> Iterator[tuple[str, int, int, list[str]]]:
+        """Stream every byte record as ``(locator, size, mtime_ns, tags)``.
+
+        One SELECT ordered by locator, then by the bytes' first sighting, with
+        each set's names in position order.
+        """
+        conn = self.model.get_connection()
+        current: tuple | None = None
+        for locator, byte_id, size, mtime_ns, name in conn.execute(_RECORDS):
+            if current is None or current[0] != byte_id:
+                if current is not None:
+                    yield current[1]
+                current = (byte_id, (locator, size, mtime_ns, []))
+            if name is not None:
+                current[1][3].append(name)
+        if current is not None:
+            yield current[1]

database/category_import.py

@@ -0,0 +1,198 @@
+"""Atomic replacement of one imported category (#221, #249 System S10).
+
+The LoRA reference import used to delete its whole category and then save
+each record in its own transaction, so a failure part way left the previous
+collection deleted and the new one partial. ``replace_category_prompts``
+does both in one ``BEGIN IMMEDIATE`` transaction: the category's current
+prompts (with their tags and image links) are replaced by the staged
+records, or, on any error, nothing changes.
+
+A staged record is ``{text, hash, tags, images}``. A record whose hash a
+prompt outside the category already has links its images to that prompt
+(the legacy import did the same); a second record with the hash of an
+earlier one links its images to that one. Neither is a new prompt.
+"""
+
+from typing import Any, Dict, List, Optional, Sequence
+
+try:
+    from .maintenance import PlanChanged
+    from .pairs import PairDependencyError
+    from .workflow_documents import documents_of_rows, release_documents
+except ImportError:  # pragma: no cover - flat test imports
+    from database.maintenance import PlanChanged
+    from database.pairs import PairDependencyError
+    from database.workflow_documents import documents_of_rows, release_documents
+
+
+class EmptyReplacement(ValueError):
+    """The replacement has no record and would remove every current import."""
+
+
+# The category's prompt ids, as a subquery (one bound parameter).
+_IN_CATEGORY = "SELECT id FROM prompts WHERE category = ?"
+
+# Each record's outcome and the result counter it adds to.
+_COUNTERS = {"imported": "imported", "linked": "linked", "duplicate": "duplicates"}
+
+
+def _placeholders(values: Sequence[Any]) -> str:
+    return ",".join("?" * len(values))
+
+
+class CategoryImportOperations:
+    """Read the state of an imported category and replace it atomically."""
+
+    def category_import_state(self, category: str) -> Dict[str, Any]:
+        """``{ids, paired}``: the category's prompt ids (ascending) and how
+        many of them a saved pair uses (those block a replacement)."""
+        with self.model.get_connection() as conn:
+            ids = [
+                row[0]
+                for row in conn.execute(
+                    "SELECT id FROM prompts WHERE category = ? ORDER BY id",
+                    (category,),
+                ).fetchall()
+            ]
+            paired = conn.execute(
+                "SELECT COUNT(*) FROM prompts WHERE category = ? AND id IN "
+                "(SELECT positive_prompt_id FROM prompt_pairs UNION "
+                "SELECT negative_prompt_id FROM prompt_pairs)",
+                (category,),
+            ).fetchone()[0]
+        return {"ids": ids, "paired": paired}
+
+    def existing_prompt_hashes(
+        self, hashes: Sequence[str], exclude_category: str
+    ) -> set:
+        """The given hashes a prompt outside ``exclude_category`` already has."""
+        found = set()
+        wanted = list(dict.fromkeys(hashes))
+        with self.model.get_connection() as conn:
+            for start in range(0, len(wanted), 500):
+                chunk = wanted[start : start + 500]
+                rows = conn.execute(
+                    f"SELECT hash FROM prompts WHERE hash IN ({_placeholders(chunk)}) "
+                    "AND (category IS NULL OR category != ?)",
+                    [*chunk, exclude_category],
+                ).fetchall()
+                found.update(row[0] for row in rows)
+        return found
+
+    def replace_category_prompts(
+        self,
+        category: str,
+        records: List[Dict[str, Any]],
+        *,
+        expected_ids: Optional[List[int]] = None,
+        allow_empty: bool = False,
+    ) -> Dict[str, Any]:
+        """Replace every prompt of ``category`` with ``records`` in one transaction.
+
+        Raises (and changes nothing): ``PlanChanged`` when ``expected_ids`` is
+        given and the category's prompt ids differ from it;
+        ``PairDependencyError`` when a saved pair uses one of its prompts;
+        ``EmptyReplacement`` when ``records`` is empty while the category has
+        prompts, unless ``allow_empty`` (an empty replacement that was
+        reviewed). Any other error rolls everything back too.
+
+        Returns ``{removed, imported, linked, duplicates, images, rows}``:
+        prompts removed and created, records linked to an existing prompt or
+        to an earlier record with the same text, image links added, and per
+        record ``{name, outcome}`` (``imported``/``linked``/``duplicate``).
+        """
+        with self._pair_transaction() as conn:
+            ids = [
+                row[0]
+                for row in conn.execute(
+                    "SELECT id FROM prompts WHERE category = ? ORDER BY id",
+                    (category,),
+                ).fetchall()
+            ]
+            if expected_ids is not None and ids != list(expected_ids):
+                raise PlanChanged("The imported prompts changed since the review")
+            if not records and ids and not allow_empty:
+                raise EmptyReplacement(
+                    "Nothing to import; the previous imports were kept"
+                )
+            self._ensure_category_unpaired(conn, category)
+            removed = self._delete_category_rows(conn, category, ids)
+            result = {
+                "removed": removed,
+                "imported": 0,
+                "linked": 0,
+                "duplicates": 0,
+                "images": 0,
+                "rows": [],
+            }
+            created = {}
+            for record in records:
+                self._import_record(conn, category, record, created, result)
+            return result
+
+    def _delete_category_rows(self, conn, category: str, ids: List[int]) -> int:
+        """Delete the category's prompts with their image links and tags.
+
+        No commit. The rows are selected by category (just verified against
+        ``ids`` under the write lock), so no statement binds one parameter per
+        prompt and SQLite's variable limit never caps a collection.
+        """
+        if not ids:
+            return 0
+        documents = documents_of_rows(conn, "prompt_id", ids)  # chunked
+        conn.execute(
+            f"DELETE FROM generated_images WHERE prompt_id IN ({_IN_CATEGORY})",
+            (category,),
+        )
+        release_documents(conn, documents)
+        conn.execute(
+            f"DELETE FROM prompt_tags WHERE prompt_id IN ({_IN_CATEGORY})",
+            (category,),
+        )
+        return conn.execute(
+            "DELETE FROM prompts WHERE category = ?", (category,)
+        ).rowcount
+
+    @staticmethod
+    def _ensure_category_unpaired(conn, category: str) -> None:
+        """Raise PairDependencyError when a saved pair uses a category prompt."""
+        row = conn.execute(
+            f"SELECT id FROM prompt_pairs WHERE positive_prompt_id IN ({_IN_CATEGORY}) "
+            f"OR negative_prompt_id IN ({_IN_CATEGORY}) LIMIT 1",
+            (category, category),
+        ).fetchone()
+        if row:
+            raise PairDependencyError(
+                f"Prompt belongs to saved pair {row[0]}; delete the pair first "
+                "or save changed text as a new prompt"
+            )
+
+    def _import_record(self, conn, category, record, created, result) -> None:
+        """Save one staged record on ``conn``, or link it to the prompt it repeats."""
+        prompt_hash = record["hash"]
+        name = record.get("name") or record["text"]
+        if prompt_hash in created:
+            prompt_id, outcome = created[prompt_hash], "duplicate"
+        else:
+            row = conn.execute(
+                "SELECT id FROM prompts WHERE hash = ?", (prompt_hash,)
+            ).fetchone()
+            if row is not None:
+                prompt_id, outcome = row[0], "linked"
+            else:
+                prompt_id = self._insert_prompt(
+                    conn,
+                    record["text"],
+                    category,
+                    list(record.get("tags") or []),
+                    None,
+                    None,
+                    prompt_hash,
+                )
+                outcome = "imported"
+            created[prompt_hash] = prompt_id
+        result[_COUNTERS[outcome]] += 1
+        for image in record.get("images") or []:
+            if self._insert_image_link(conn, prompt_id, image):
+                result["images"] += 1
+        result["rows"].append({"name": name, "outcome": outcome})

database/duplicate_copies.py

@@ -0,0 +1,215 @@
+"""Records of duplicate image files: count them and move them (#249 S6, T17).
+
+A removed copy's records move to its verified kept copy before the copy is
+unlinked (``py/api/duplicate_files.py``), one file per short ``BEGIN
+IMMEDIATE`` transaction, so no row ever names a deleted file:
+
+- its image records (``generated_images``), found under every spelling the
+  file is linked by. A record moves to the kept copy, unless its prompt
+  already links the kept copy: then its saved-pair links join that record
+  and it is deleted;
+- its favorite;
+- its image tags (``media_image_tags`` of its byte record), joined onto the
+  kept copy's byte record as their union, in order;
+- its pending Needs review items, settled: only an item still ``pending`` at
+  revision 0 with no decision row is removed (``review._SETTLE_ITEM``), so a
+  decided item is never touched.
+
+Every statement is an index search (see ``FILE_LINK_SQL``).
+"""
+
+import os
+
+from .annotation_schema import BYTES_TABLE, FILES_TABLE
+from .annotations import _BUMP_INDEX
+from .models import normalize_image_path, utc_now_iso
+from .review import _BUMP_REVIEW_INDEX, _PENDING_OF_LOCATOR, _SETTLE_ITEM
+from .review_changes import notify_review_change
+from .workflow_documents import documents_of_rows, release_documents
+
+LINKS_BY_FILE_SQL = "SELECT id, prompt_id FROM generated_images WHERE file_path = ?"
+LEGACY_LINKS_SQL = (
+    "SELECT id, prompt_id FROM generated_images "
+    "WHERE image_path = ? AND file_path IS NULL"
+)
+KEEPER_LINK_SQL = (
+    "SELECT id FROM generated_images WHERE prompt_id = ? AND file_path = ?"
+)
+MOVE_PAIR_LINKS_SQL = (
+    "UPDATE OR IGNORE image_prompt_pairs SET image_id = ? WHERE image_id = ?"
+)
+MOVE_LINK_SQL = (
+    "UPDATE generated_images SET image_path = ?, filename = ?, file_path = ? "
+    "WHERE id = ?"
+)
+DELETE_LINK_SQL = "DELETE FROM generated_images WHERE id = ?"
+FAVORITE_SQL = (
+    "SELECT created_at FROM image_favorites WHERE root_key = ? AND relative_path = ?"
+)
+INSERT_FAVORITE_SQL = "INSERT OR IGNORE INTO image_favorites VALUES (?, ?, ?)"
+DELETE_FAVORITE_SQL = (
+    "DELETE FROM image_favorites WHERE root_key = ? AND relative_path = ?"
+)
+COPY_BYTE_SQL = (
+    f"SELECT b.id FROM {FILES_TABLE} f JOIN {BYTES_TABLE} b "  # nosec B608
+    "ON b.file_id=f.id WHERE f.locator=? AND b.sha256=?"
+)
+COPY_TAG_COUNT_SQL = (
+    "SELECT COUNT(*) FROM media_image_tags WHERE byte_id=?"  # nosec B608
+)
+PENDING_COUNT_SQL = (
+    f"SELECT COUNT(*) FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    "JOIN media_review_items i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND +i.state='pending' AND i.revision=0 "
+    "AND NOT EXISTS (SELECT 1 FROM media_review_decisions d WHERE d.item_id=i.id)"
+)
+FILE_LINK_SQL = (
+    LINKS_BY_FILE_SQL,
+    LEGACY_LINKS_SQL,
+    KEEPER_LINK_SQL,
+    MOVE_PAIR_LINKS_SQL,
+    MOVE_LINK_SQL,
+    DELETE_LINK_SQL,
+    FAVORITE_SQL,
+    DELETE_FAVORITE_SQL,
+    COPY_BYTE_SQL,
+    COPY_TAG_COUNT_SQL,
+    PENDING_COUNT_SQL,
+)
+
+
+def _file_link_rows(conn, spelled):
+    """``(id, prompt_id)`` of the image records linking one spelling."""
+    rows = list(conn.execute(LINKS_BY_FILE_SQL, (normalize_image_path(spelled),)))
+    rows += list(conn.execute(LEGACY_LINKS_SQL, (spelled,)))
+    return [tuple(row) for row in rows]
+
+
+def _copy_byte(conn, locator, sha256):
+    row = conn.execute(COPY_BYTE_SQL, (locator, sha256)).fetchone()
+    return row[0] if row else None
+
+
+class DuplicateCopyOperations:
+    """PromptDatabase mixin: the records of duplicate copies."""
+
+    def count_copy_records(self, copies) -> dict:
+        """``{path: {links, tags, review}}`` for each copy (one read snapshot).
+
+        ``copies`` maps each path to ``{spellings, locator, sha256}``:
+        ``links`` counts the image records under any spelling, ``tags`` the
+        image tags of its bytes and ``review`` its pending, undecided Needs
+        review items.
+        """
+        from .maintenance import _read_snapshot
+
+        def read(conn):
+            found = {}
+            for path, copy in copies.items():
+                ids = {
+                    row[0]
+                    for spelled in copy["spellings"]
+                    for row in _file_link_rows(conn, spelled)
+                }
+                byte_id = _copy_byte(conn, copy["locator"], copy["sha256"])
+                tags = (
+                    conn.execute(COPY_TAG_COUNT_SQL, (byte_id,)).fetchone()[0]
+                    if byte_id
+                    else 0
+                )
+                review = conn.execute(PENDING_COUNT_SQL, (copy["locator"],))
+                found[path] = {
+                    "links": len(ids),
+                    "tags": tags,
+                    "review": review.fetchone()[0],
+                }
+            return found
+
+        return _read_snapshot(self.model.db_path, read)
+
+    def move_copy_records(self, move) -> dict:
+        """Move one removed copy's records to its kept copy (one transaction).
+
+        ``move``: ``removed`` and ``kept`` (the spellings each file is linked
+        by), ``targets`` (``{removed spelling: kept spelling}``, the spelling
+        a moved record takes; else ``kept_path``), ``favorites`` (``(root key,
+        removed key, kept root key, kept key)`` per listed root holding the
+        file), ``removed_locator``,
+        ``kept_locator``, ``sha256`` and the kept copy's ``kept_size`` and
+        ``kept_mtime_ns``. Returns ``{moved, merged, favorites, tags,
+        review}``.
+        """
+        counts = dict.fromkeys(("moved", "merged", "favorites", "tags", "review"), 0)
+        kept_keys = [normalize_image_path(path) for path in move["kept"]]
+        with self._pair_transaction() as conn:
+            for spelled in dict.fromkeys(move["removed"]):
+                target = move["targets"].get(spelled, move["kept_path"])
+                keys = list(dict.fromkeys(kept_keys + [normalize_image_path(target)]))
+                for image_id, prompt_id in _file_link_rows(conn, spelled):
+                    existing = None
+                    for key in keys:
+                        existing = conn.execute(
+                            KEEPER_LINK_SQL, (prompt_id, key)
+                        ).fetchone()
+                        if existing is not None:
+                            break
+                    if existing is not None:
+                        conn.execute(MOVE_PAIR_LINKS_SQL, (existing[0], image_id))
+                        documents = documents_of_rows(conn, "id", [image_id])
+                        conn.execute(DELETE_LINK_SQL, (image_id,))
+                        release_documents(conn, documents)
+                        counts["merged"] += 1
+                        continue
+                    conn.execute(
+                        MOVE_LINK_SQL,
+                        (
+                            target,
+                            os.path.basename(target),
+                            normalize_image_path(target),
+                            image_id,
+                        ),
+                    )
+                    counts["moved"] += 1
+            for favorite in move.get("favorites", ()):
+                counts["favorites"] += _move_favorite(conn, *favorite)
+            counts["tags"] = self._move_copy_tags(conn, move)
+            for (item_id,) in conn.execute(
+                _PENDING_OF_LOCATOR, (move["removed_locator"],)
+            ).fetchall():
+                counts["review"] += conn.execute(_SETTLE_ITEM, (item_id,)).rowcount
+            if counts["review"]:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        if counts["review"]:
+            notify_review_change(self.model.db_path)
+        return counts
+
+    def _move_copy_tags(self, conn, move):
+        """Join the removed copy's image tags onto the kept copy's; the count added."""
+        removed = _copy_byte(conn, move["removed_locator"], move["sha256"])
+        if removed is None:
+            return 0
+        tags = self._byte_tags(conn, removed)
+        if not tags:
+            return 0
+        now = utc_now_iso()
+        file_id = self._annotation_owner(conn, move["kept_locator"], now)
+        byte_id, _revision, created, moved = self._annotation_byte(
+            conn, file_id, move["sha256"], move["kept_size"], move["kept_mtime_ns"], now
+        )
+        current = self._byte_tags(conn, byte_id)
+        union = current + [tag for tag in tags if tag not in current]
+        if union != current:
+            self._write_image_tags(conn, byte_id, union)
+        elif created or moved:
+            conn.execute(_BUMP_INDEX)
+        return len(union) - len(current)
+
+
+def _move_favorite(conn, root_key, removed_key, kept_root, kept_key):
+    row = conn.execute(FAVORITE_SQL, (root_key, removed_key)).fetchone()
+    if row is None:
+        return 0
+    conn.execute(INSERT_FAVORITE_SQL, (kept_root, kept_key, row[0]))
+    conn.execute(DELETE_FAVORITE_SQL, (root_key, removed_key))
+    return 1

database/execution_context_schema.py

@@ -0,0 +1,406 @@
+"""Owned execution-context v1 tables within execution-history schema v2.
+
+Content IDs hash exact scalar values; evidence JSON contains metadata only.
+All validation scans use cursors and bounded per-row payloads.
+"""
+
+import json
+import sqlite3
+
+from .execution_history_schema import _compact, _uuid_check
+
+try:
+    from ..utils.execution_contexts import (
+        BUNDLE_STATES,
+        CAPTURE_REASONS,
+        EVIDENCE_SOURCES,
+        UNKNOWN_REASONS,
+        MAX_CONTEXT_BYTES,
+        ContextBundle,
+        PromptSnapshot,
+        RecipeContext,
+        TextEvidence,
+        validate_bundle,
+        content_digest,
+        text_size,
+    )
+except ImportError:
+    from utils.execution_contexts import (
+        BUNDLE_STATES,
+        CAPTURE_REASONS,
+        EVIDENCE_SOURCES,
+        UNKNOWN_REASONS,
+        MAX_CONTEXT_BYTES,
+        ContextBundle,
+        PromptSnapshot,
+        RecipeContext,
+        TextEvidence,
+        validate_bundle,
+        content_digest,
+        text_size,
+    )
+
+
+FIELDS = ("raw", "effective", "prepend", "append")
+TABLE_COLUMNS = {
+    "execution_context_texts": (
+        ("text_id", "TEXT", 1, 1),
+        ("text_value", "TEXT", 1, 0),
+    ),
+    "execution_prompt_snapshots": (
+        ("snapshot_id", "TEXT", 1, 1),
+        *((f"{field}_text_id", "TEXT", 0, 0) for field in FIELDS),
+    ),
+    "execution_contexts": (
+        ("context_id", "TEXT", 1, 1),
+        ("positive_snapshot_id", "TEXT", 0, 0),
+        ("negative_snapshot_id", "TEXT", 0, 0),
+    ),
+    "execution_run_contexts": (
+        ("run_id", "TEXT", 1, 1),
+        ("anchor_id", "TEXT", 1, 2),
+        ("context_id", "TEXT", 1, 0),
+        ("evidence_json", "TEXT", 1, 0),
+        ("positive_prompt_id", "INTEGER", 0, 0),
+        ("negative_prompt_id", "INTEGER", 0, 0),
+        ("pair_id", "INTEGER", 0, 0),
+        ("reference_epoch", "TEXT", 0, 0),
+        ("capture_version", "INTEGER", 1, 0),
+    ),
+    "execution_context_capture": (
+        ("run_id", "TEXT", 1, 1),
+        ("state", "TEXT", 1, 0),
+        ("reason", "TEXT", 1, 0),
+        ("format_version", "INTEGER", 1, 0),
+    ),
+}
+
+
+def _digest_check(column: str, nullable: bool = False) -> str:
+    check = (
+        f"typeof({column}) = 'text' AND length({column}) = 64 "
+        f"AND length(CAST({column} AS BLOB)) = 64 "
+        f"AND {column} NOT GLOB '*[^0-9a-f]*'"
+    )
+    return f"{column} IS NULL OR ({check})" if nullable else check
+
+
+def _bounded_uuid_check(column: str) -> str:
+    return f"{_uuid_check(column)} AND length(CAST({column} AS BLOB)) = 36"
+
+
+def _enum_check(column: str, values: frozenset[str]) -> str:
+    return (
+        f"typeof({column}) = 'text' AND {column} IN ("
+        + ",".join("'" + value + "'" for value in sorted(values))
+        + ")"
+    )
+
+
+TABLE_CHECKS = {
+    "execution_context_texts": (
+        _digest_check("text_id"),
+        "typeof(text_value) = 'text' AND length(CAST(text_value AS BLOB)) <= 65536",
+    ),
+    "execution_prompt_snapshots": (
+        _digest_check("snapshot_id"),
+        *(_digest_check(f"{field}_text_id", True) for field in FIELDS),
+    ),
+    "execution_contexts": (
+        _digest_check("context_id"),
+        _digest_check("positive_snapshot_id", True),
+        _digest_check("negative_snapshot_id", True),
+        "positive_snapshot_id IS NOT NULL OR negative_snapshot_id IS NOT NULL",
+    ),
+    "execution_run_contexts": (
+        _bounded_uuid_check("run_id"),
+        "typeof(anchor_id) = 'text' AND length(anchor_id) BETWEEN 1 AND 128 "
+        "AND length(CAST(anchor_id AS BLOB)) <= 512",
+        _digest_check("context_id"),
+        "typeof(evidence_json) = 'text' AND length(CAST(evidence_json AS "
+        "BLOB)) <= 8192",
+        *(
+            f"{column} IS NULL OR (typeof({column}) = 'integer' AND {column} > 0)"
+            for column in ("positive_prompt_id", "negative_prompt_id", "pair_id")
+        ),
+        f"reference_epoch IS NULL OR ({_bounded_uuid_check('reference_epoch')})",
+        "typeof(capture_version) = 'integer' AND capture_version = 1",
+    ),
+    "execution_context_capture": (
+        _bounded_uuid_check("run_id"),
+        _enum_check("state", BUNDLE_STATES),
+        _enum_check("reason", CAPTURE_REASONS),
+        "length(CAST(reason AS BLOB)) <= 64",
+        "typeof(format_version) = 'integer' AND format_version = 1",
+    ),
+}
+# (source column, destination table, destination column); actions always NO ACTION.
+TABLE_FOREIGN_KEYS = {
+    "execution_context_texts": (),
+    "execution_prompt_snapshots": tuple(
+        (f"{field}_text_id", "execution_context_texts", "text_id") for field in FIELDS
+    ),
+    "execution_contexts": tuple(
+        (f"{role}_snapshot_id", "execution_prompt_snapshots", "snapshot_id")
+        for role in ("positive", "negative")
+    ),
+    "execution_run_contexts": (
+        ("run_id", "execution_runs", "run_id"),
+        ("context_id", "execution_contexts", "context_id"),
+    ),
+    "execution_context_capture": (("run_id", "execution_runs", "run_id"),),
+}
+# Only SQLite-generated primary-key indexes are unique. No secondary indexes required.
+
+
+def install_context_schema(conn: sqlite3.Connection) -> None:
+    """Install tables inside the caller's transaction, without committing."""
+    for table, columns in TABLE_COLUMNS.items():
+        definitions = [
+            f"{name} {kind}{' NOT NULL' if required else ''}"
+            for name, kind, required, _ in columns
+        ]
+        primary = [name for name, _, _, order in columns if order]
+        definitions.append(f"PRIMARY KEY ({', '.join(primary)})")
+        definitions.extend(f"CHECK({check})" for check in TABLE_CHECKS[table])
+        definitions.extend(
+            f"FOREIGN KEY ({column}) REFERENCES {target}({key}) "
+            "ON UPDATE NO ACTION ON DELETE NO ACTION"
+            for column, target, key in TABLE_FOREIGN_KEYS[table]
+        )
+        conn.execute(f"CREATE TABLE {table} ({', '.join(definitions)})")
+
+
+def _unique_object(pairs: list[tuple[str, object]]) -> dict:
+    result = {}
+    for key, value in pairs:
+        if key in result:
+            raise ValueError("Duplicate evidence key")
+        result[key] = value
+    return result
+
+
+def _snapshot_values(conn: sqlite3.Connection, snapshot_id: str) -> tuple:
+    row = conn.execute(
+        "SELECT raw_text_id,effective_text_id,prepend_text_id,append_text_id "
+        "FROM execution_prompt_snapshots WHERE snapshot_id=?",
+        (snapshot_id,),
+    ).fetchone()
+    if row is None:
+        raise ValueError("Missing snapshot")
+    values = []
+    for text_id in row:
+        if text_id is None:
+            values.append(None)
+        else:
+            text = conn.execute(
+                "SELECT text_value FROM execution_context_texts WHERE text_id=?",
+                (text_id,),
+            ).fetchone()
+            if text is None:
+                raise ValueError("Missing context text")
+            text_size(text[0])
+            values.append(text[0])
+    return tuple(values)
+
+
+def _validate_evidence(
+    conn: sqlite3.Connection, anchor_id: str, context_id: str, encoded: str
+) -> tuple[RecipeContext, bool]:
+    text_size(encoded, 8192)
+    evidence = json.loads(encoded, object_pairs_hook=_unique_object)
+    if (
+        type(evidence) is not dict
+        or set(evidence)
+        != {"format_version", "role_provenance", "positive", "negative"}
+        or type(evidence["format_version"]) is not int
+        or evidence["format_version"] != 1
+        or evidence["role_provenance"] != "inferred"
+    ):
+        raise ValueError("Invalid context evidence")
+    row = conn.execute(
+        "SELECT positive_snapshot_id,negative_snapshot_id FROM "
+        "execution_contexts WHERE context_id=?",
+        (context_id,),
+    ).fetchone()
+    if row is None:
+        raise ValueError("Missing context")
+    complete = True
+    snapshots = []
+    for role, snapshot_id in zip(("positive", "negative"), row):
+        metadata = evidence[role]
+        if snapshot_id is None:
+            if metadata is not None:
+                raise ValueError("Unexpected role evidence")
+            complete = False
+            snapshots.append(None)
+            continue
+        values = _snapshot_values(conn, snapshot_id)
+        if type(metadata) is not dict or set(metadata) != set(FIELDS):
+            raise ValueError("Invalid snapshot evidence")
+        fields = []
+        for field, value in zip(FIELDS, values):
+            item = metadata[field]
+            if type(item) is not dict or set(item) != {
+                "provenance",
+                "source",
+                "reason",
+            }:
+                raise ValueError("Invalid field evidence")
+            provenance, source, reason = (
+                item["provenance"],
+                item["source"],
+                item["reason"],
+            )
+            if (
+                type(provenance) is not str
+                or type(source) is not str
+                or source not in EVIDENCE_SOURCES.get(provenance, ())
+            ):
+                raise ValueError("Invalid evidence source")
+            if value is None:
+                complete = False
+                if (
+                    provenance != "unknown"
+                    or type(reason) is not str
+                    or reason not in UNKNOWN_REASONS
+                ):
+                    raise ValueError("Unknown evidence requires reason")
+            elif provenance == "unknown" or reason is not None:
+                raise ValueError("Known evidence metadata mismatch")
+            fields.append(TextEvidence(value, provenance, source, reason))
+        snapshots.append(PromptSnapshot(*fields))
+    return RecipeContext(anchor_id, *snapshots, evidence["role_provenance"]), complete
+
+
+def context_schema_error(conn: sqlite3.Connection) -> str | None:
+    """Validate exact owned schema, bounded evidence, fingerprints and links."""
+    if conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type IN ('trigger','view') LIMIT 1"
+    ).fetchone():
+        return "Unsupported execution context trigger/view"
+    tables = {
+        row[0]: row[1]
+        for row in conn.execute("SELECT name,sql FROM sqlite_master WHERE type='table'")
+        if row[0] in TABLE_COLUMNS
+    }
+    if tables.keys() != TABLE_COLUMNS.keys():
+        return "Incomplete execution context schema"
+    for table, expected in TABLE_COLUMNS.items():
+        columns = tuple(
+            (row[1], row[2].upper(), row[3], row[5])
+            for row in conn.execute(f"PRAGMA table_xinfo({table})")
+        )
+        if columns != expected:
+            return f"{table} has invalid execution context columns"
+        sql = _compact(tables[table])
+        checks = TABLE_CHECKS[table]
+        if sql.count("check(") != len(checks) or any(
+            f"check({_compact(check)})" not in sql for check in checks
+        ):
+            return f"{table} has invalid execution context checks"
+        foreign = tuple(
+            sorted(
+                (row[3], row[2], row[4], row[5], row[6], row[7])
+                for row in conn.execute(f"PRAGMA foreign_key_list({table})")
+            )
+        )
+        expected_foreign = tuple(
+            sorted(
+                (*key, "NO ACTION", "NO ACTION", "NONE")
+                for key in TABLE_FOREIGN_KEYS[table]
+            )
+        )
+        if foreign != expected_foreign:
+            return f"{table} has invalid execution context foreign keys"
+        for index in conn.execute(f"PRAGMA index_list({table})"):
+            if not index[2]:
+                continue
+            name = index[1].replace('"', '""')
+            keys = tuple(
+                row for row in conn.execute(f'PRAGMA index_xinfo("{name}")') if row[5]
+            )
+            primary = tuple(column[0] for column in expected if column[3])
+            if (
+                index[3] != "pk"
+                or index[4]
+                or tuple(key[2] for key in keys) != primary
+                or any(key[3] or key[4] != "BINARY" for key in keys)
+            ):
+                return f"{table} has unexpected execution context uniqueness"
+        predicate = " AND ".join(f"({check})" for check in checks)
+        if conn.execute(
+            f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1"
+        ).fetchone():
+            return f"{table} contains invalid execution context records"
+        if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
+            return f"{table} contains orphan execution context records"
+    if conn.execute(
+        "SELECT 1 FROM execution_run_contexts GROUP BY run_id HAVING "
+        "COUNT(*) > 16 LIMIT 1"
+    ).fetchone():
+        return "Too many execution contexts per run"
+    try:
+        for text_id, text in conn.execute(
+            "SELECT text_id,text_value FROM execution_context_texts"
+        ):
+            text_size(text)
+            if content_digest("text", (text,)) != text_id:
+                raise ValueError("Invalid text fingerprint")
+        for (snapshot_id,) in conn.execute(
+            "SELECT snapshot_id FROM execution_prompt_snapshots"
+        ):
+            if (
+                content_digest("snapshot", _snapshot_values(conn, snapshot_id))
+                != snapshot_id
+            ):
+                raise ValueError("Invalid snapshot fingerprint")
+        for context_id, positive, negative in conn.execute(
+            "SELECT context_id,positive_snapshot_id,negative_snapshot_id FROM "
+            "execution_contexts"
+        ):
+            if content_digest("context", (positive, negative)) != context_id:
+                raise ValueError("Invalid context fingerprint")
+            payload_size = 0
+            for snapshot_id in (positive, negative):
+                if snapshot_id is not None:
+                    payload_size += sum(
+                        text_size(value)
+                        for value in _snapshot_values(conn, snapshot_id)
+                        if value is not None
+                    )
+            if payload_size > MAX_CONTEXT_BYTES:
+                raise ValueError("Oversized detached context")
+        previous_run = None
+        contexts = ()
+        for run_id, anchor_id, context_id, evidence in conn.execute(
+            "SELECT run_id,anchor_id,context_id,evidence_json FROM "
+            "execution_run_contexts ORDER BY run_id,anchor_id"
+        ):
+            if run_id != previous_run:
+                contexts = ()
+                previous_run = run_id
+            context, complete = _validate_evidence(
+                conn, anchor_id, context_id, evidence
+            )
+            status = conn.execute(
+                "SELECT state,reason FROM execution_context_capture WHERE run_id=?",
+                (run_id,),
+            ).fetchone()
+            if status is None:
+                raise ValueError("Context link has no capture header")
+            state, reason = status
+            if state == "complete" and not complete:
+                raise ValueError("Incomplete evidence marked complete")
+            candidate = (*contexts, context)
+            validate_bundle(ContextBundle(candidate, state, reason))
+            contexts = candidate
+        if conn.execute(
+            "SELECT 1 FROM execution_context_capture c WHERE state='complete' "
+            "AND NOT EXISTS (SELECT 1 FROM execution_run_contexts r WHERE "
+            "r.run_id=c.run_id) LIMIT 1"
+        ).fetchone():
+            raise ValueError("Complete capture has no contexts")
+    except (ValueError, TypeError, RecursionError, OverflowError) as exc:
+        return f"Invalid execution context content: {type(exc).__name__}"
+    return None

database/execution_contexts.py

@@ -0,0 +1,411 @@
+"""Immutable execution capture and exact, epoch-fenced library navigation hints."""
+
+from contextlib import ExitStack
+import json
+import sqlite3
+from typing import Callable, ContextManager, Iterator
+
+try:
+    from ..utils.execution_history_events import RunFinish, encode_command
+    from ..utils.execution_contexts import (
+        CAPTURE_REASONS,
+        ContextBundle,
+        content_digest,
+        validate_bundle,
+    )
+except ImportError:
+    from utils.execution_history_events import RunFinish, encode_command
+    from utils.execution_contexts import (
+        CAPTURE_REASONS,
+        ContextBundle,
+        content_digest,
+        validate_bundle,
+    )
+
+
+FIELDS = ("raw", "effective", "prepend", "append")
+ROLES = ("positive", "negative")
+
+
+def _prepare(bundle: ContextBundle) -> list[dict]:
+    validate_bundle(bundle)
+    if bundle.reason not in CAPTURE_REASONS:
+        raise ValueError("Invalid capture reason")
+    prepared = []
+    complete = bool(bundle.contexts)
+    for context in bundle.contexts:
+        evidence = {"format_version": 1, "role_provenance": context.role_provenance}
+        snapshots = []
+        for role in ROLES:
+            snapshot = getattr(context, role)
+            if snapshot is None:
+                snapshots.append(None)
+                evidence[role] = None
+                complete = False
+                continue
+            values = tuple(getattr(snapshot, field).value for field in FIELDS)
+            complete = complete and all(value is not None for value in values)
+            snapshots.append((content_digest("snapshot", values), values))
+            evidence[role] = {
+                field: {
+                    "provenance": getattr(snapshot, field).provenance,
+                    "source": getattr(snapshot, field).source,
+                    "reason": getattr(snapshot, field).reason,
+                }
+                for field in FIELDS
+            }
+        if all(snapshot is None for snapshot in snapshots):
+            raise ValueError("Context must contain a snapshot")
+        context_id = content_digest(
+            "context", tuple(s[0] if s else None for s in snapshots)
+        )
+        encoded = json.dumps(
+            evidence,
+            ensure_ascii=False,
+            allow_nan=False,
+            separators=(",", ":"),
+            sort_keys=True,
+        )
+        if len(encoded.encode("utf-8")) > 8192:
+            raise ValueError("Oversized context evidence")
+        prepared.append(
+            {
+                "anchor_id": context.anchor_id,
+                "context_id": context_id,
+                "snapshots": snapshots,
+                "evidence_json": encoded,
+            }
+        )
+    if bundle.state == "complete" and not complete:
+        raise ValueError("Incomplete scalar capture")
+    return sorted(prepared, key=lambda item: item["anchor_id"])
+
+
+def _existing_row(
+    conn: sqlite3.Connection, table: str, columns: tuple[str, ...], values: tuple
+) -> bool:
+    row = conn.execute(
+        f"SELECT {','.join(columns[1:])} FROM {table} WHERE {columns[0]}=?",
+        (values[0],),
+    ).fetchone()
+    if row is None:
+        return False
+    if tuple(row) != values[1:]:
+        raise ValueError("Execution context digest collision")
+    return True
+
+
+def _content_rows(item: dict) -> Iterator[tuple[str, tuple[str, ...], tuple]]:
+    snapshot_ids = []
+    for snapshot in item["snapshots"]:
+        if snapshot is None:
+            snapshot_ids.append(None)
+            continue
+        snapshot_id, values = snapshot
+        text_ids = []
+        for value in values:
+            text_id = None if value is None else content_digest("text", (value,))
+            if value is not None:
+                yield "execution_context_texts", ("text_id", "text_value"), (
+                    text_id,
+                    value,
+                )
+            text_ids.append(text_id)
+        yield "execution_prompt_snapshots", (
+            "snapshot_id",
+            *(f"{field}_text_id" for field in FIELDS),
+        ), (snapshot_id, *text_ids)
+        snapshot_ids.append(snapshot_id)
+    yield "execution_contexts", (
+        "context_id",
+        "positive_snapshot_id",
+        "negative_snapshot_id",
+    ), (item["context_id"], *snapshot_ids)
+
+
+def _insert_exact(
+    conn: sqlite3.Connection, table: str, columns: tuple[str, ...], values: tuple
+) -> None:
+    if not _existing_row(conn, table, columns, values):
+        names = ",".join(columns)
+        placeholders = ",".join("?" for _ in columns)
+        conn.execute(f"INSERT INTO {table} ({names}) VALUES ({placeholders})", values)
+
+
+def _role_hints(role: str) -> str:
+    return (
+        f"SELECT r.{role}_prompt_id, tx.text_value "
+        "FROM execution_run_contexts r "
+        "JOIN execution_contexts c ON c.context_id = r.context_id "
+        f"JOIN execution_prompt_snapshots s ON s.snapshot_id = c.{role}_snapshot_id "
+        "JOIN execution_context_texts tx ON tx.text_id = s.raw_text_id "
+        f"WHERE r.{role}_prompt_id IS NOT NULL AND r.reference_epoch = ("
+        "SELECT storage_epoch FROM execution_history_metadata WHERE singleton_id = 1)"
+    )
+
+
+# Set form of the read-time prompt-hint rule in get_execution_contexts():
+# a stored role hint is live only in the current storage epoch, and only when
+# its exact raw snapshot text matches exactly one prompt, which is the hinted
+# one (_prompt_hint). Evaluated once per statement, never per prompt row.
+# Defines the CTE ``live_hint_prompts(prompt_id)``; prefix it with WITH.
+LIVE_HINT_CTES = (
+    "stored_hints(prompt_id, raw) AS ("
+    + _role_hints("positive")
+    + " UNION "
+    + _role_hints("negative")
+    + "), unique_hint_texts(prompt_id, raw) AS ("
+    "SELECT MIN(id), text COLLATE BINARY FROM prompts "
+    "WHERE text COLLATE BINARY IN (SELECT raw FROM stored_hints) "
+    "GROUP BY text COLLATE BINARY HAVING COUNT(*) = 1"
+    "), live_hint_prompts(prompt_id) AS ("
+    "SELECT h.prompt_id FROM stored_hints h JOIN unique_hint_texts u "
+    "ON u.prompt_id = h.prompt_id AND u.raw = h.raw COLLATE BINARY)"
+)
+LIVE_HINT_PROMPT_IDS_SQL = (
+    "WITH " + LIVE_HINT_CTES + " SELECT DISTINCT prompt_id FROM live_hint_prompts"
+)
+
+
+def _prompt_hint(conn: sqlite3.Connection, raw: str | None) -> int | None:
+    if raw is None:
+        return None
+    rows = conn.execute(
+        "SELECT id FROM prompts WHERE text=? COLLATE BINARY LIMIT 2", (raw,)
+    ).fetchall()
+    return rows[0][0] if len(rows) == 1 else None
+
+
+def _hints(
+    conn: sqlite3.Connection, positive: str | None, negative: str | None
+) -> tuple:
+    pos_id, neg_id = _prompt_hint(conn, positive), _prompt_hint(conn, negative)
+    pair_id = None
+    negative_ambiguous = (
+        negative is not None
+        and neg_id is None
+        and conn.execute(
+            "SELECT 1 FROM prompts WHERE text=? COLLATE BINARY LIMIT 1", (negative,)
+        ).fetchone()
+        is not None
+    )
+    if pos_id is not None and negative is not None and not negative_ambiguous:
+        rows = conn.execute(
+            "SELECT pp.id FROM prompt_pairs pp LEFT JOIN prompts n ON "
+            "n.id=pp.negative_prompt_id "
+            "WHERE pp.positive_prompt_id=? AND ((pp.negative_prompt_id IS NULL AND "
+            "?='') "
+            "OR (pp.negative_prompt_id=? AND n.text=? COLLATE BINARY)) LIMIT 2",
+            (pos_id, negative, neg_id, negative),
+        ).fetchall()
+        if len(rows) == 1:
+            pair_id = rows[0][0]
+    return pos_id, neg_id, pair_id
+
+
+def _values(conn: sqlite3.Connection, snapshot_id: str | None) -> dict | None:
+    if snapshot_id is None:
+        return None
+    row = conn.execute(
+        "SELECT r.text_value,e.text_value,p.text_value,a.text_value "
+        "FROM execution_prompt_snapshots s "
+        "LEFT JOIN execution_context_texts r ON r.text_id=s.raw_text_id "
+        "LEFT JOIN execution_context_texts e ON e.text_id=s.effective_text_id "
+        "LEFT JOIN execution_context_texts p ON p.text_id=s.prepend_text_id "
+        "LEFT JOIN execution_context_texts a ON a.text_id=s.append_text_id "
+        "WHERE s.snapshot_id=?",
+        (snapshot_id,),
+    ).fetchone()
+    if row is None:
+        raise sqlite3.DatabaseError("Missing execution context snapshot")
+    return dict(zip(FIELDS, row))
+
+
+def _write_capture(
+    conn: sqlite3.Connection,
+    run_id: str,
+    expected_epoch: str,
+    bundle: ContextBundle,
+    prepared: list[dict],
+) -> bool:
+    capture = conn.execute(
+        "SELECT state,reason,format_version FROM execution_context_capture "
+        "WHERE run_id=?",
+        (run_id,),
+    ).fetchone()
+    if capture is not None:
+        for item in prepared:
+            for table, columns, values in _content_rows(item):
+                _existing_row(conn, table, columns, values)
+        links = conn.execute(
+            "SELECT anchor_id,context_id,evidence_json FROM "
+            "execution_run_contexts WHERE run_id=? ORDER BY anchor_id",
+            (run_id,),
+        ).fetchall()
+        expected = [
+            (item["anchor_id"], item["context_id"], item["evidence_json"])
+            for item in prepared
+        ]
+        # Compare JSON objects so serialization order never creates a conflict.
+        original = [(row[0], row[1], json.loads(row[2])) for row in links]
+        proposed = [
+            (anchor, context, json.loads(evidence))
+            for anchor, context, evidence in expected
+        ]
+        if capture[0] != "conflicting" and (
+            tuple(capture) != (bundle.state, bundle.reason, bundle.format_version)
+            or original != proposed
+        ):
+            conn.execute(
+                "UPDATE execution_context_capture SET "
+                "state='conflicting',reason='conflicting' WHERE run_id=?",
+                (run_id,),
+            )
+        return True
+    for item in prepared:
+        for table, columns, values in _content_rows(item):
+            _insert_exact(conn, table, columns, values)
+        raw = [snapshot[1][0] if snapshot else None for snapshot in item["snapshots"]]
+        positive_id, negative_id, pair_id = _hints(conn, *raw)
+        conn.execute(
+            "INSERT INTO "
+            "execution_run_contexts(run_id,anchor_id,context_id,evidence_json,"
+            "positive_prompt_id,negative_prompt_id,pair_id,reference_epoch,"
+            "capture_version) VALUES(?,?,?,?,?,?,?,?,1)",
+            (
+                run_id,
+                item["anchor_id"],
+                item["context_id"],
+                item["evidence_json"],
+                positive_id,
+                negative_id,
+                pair_id,
+                expected_epoch,
+            ),
+        )
+    conn.execute(
+        "INSERT INTO "
+        "execution_context_capture(run_id,state,reason,format_version) "
+        "VALUES(?,?,?,?)",
+        (run_id, bundle.state, bundle.reason, bundle.format_version),
+    )
+    return True
+
+
+class ExecutionContextOperations:
+    """Compose with ExecutionHistoryOperations on PromptDatabase."""
+
+    def write_execution_contexts(
+        self, run_id: str, expected_epoch: str, bundle: ContextBundle
+    ) -> bool:
+        prepared = _prepare(bundle)
+        with self._execution_transaction(expected_epoch) as conn:
+            if (
+                conn.execute(
+                    "SELECT 1 FROM execution_runs WHERE run_id=? AND storage_epoch=?",
+                    (run_id, expected_epoch),
+                ).fetchone()
+                is None
+            ):
+                return False
+            return _write_capture(conn, run_id, expected_epoch, bundle, prepared)
+
+    def write_execution_contexts_for_finish(
+        self,
+        command: RunFinish,
+        bundle: ContextBundle,
+        *,
+        generation_guard: Callable[[], ContextManager[bool]] | None = None,
+    ) -> bool:
+        encode_command(command)
+        if type(command) is not RunFinish:
+            raise ValueError("Expected execution finish command")
+        prepared = _prepare(bundle)
+        # Enter the guard after BEGIN, and keep it until COMMIT/ROLLBACK exits.
+        with ExitStack() as guards:
+            with self._execution_transaction(command.storage_epoch) as conn:
+                if generation_guard is not None and not guards.enter_context(
+                    generation_guard()
+                ):
+                    return False
+                row = conn.execute(
+                    "SELECT run_id FROM execution_runs WHERE session_id=? "
+                    "AND queue_item_id=? AND storage_epoch=?",
+                    (command.session_id, command.queue_item_id, command.storage_epoch),
+                ).fetchone()
+                if row is None:
+                    return False
+                return _write_capture(
+                    conn, row[0], command.storage_epoch, bundle, prepared
+                )
+
+    def get_execution_contexts(self, run_id: str) -> dict:
+        """Return first-capture scalars; hints are None unless currently valid.
+
+        reference_epoch remains historical evidence even when hints are invalidated.
+        Reads never rebind a first-capture reference or alter stored capture data.
+        """
+        conn = self.model.get_connection()
+        own_transaction = not conn.in_transaction
+        if own_transaction:
+            conn.execute("BEGIN")
+        try:
+            capture = conn.execute(
+                "SELECT state,reason,format_version FROM execution_context_capture "
+                "WHERE run_id=?",
+                (run_id,),
+            ).fetchone()
+            if capture is None:
+                return {
+                    "state": "not_captured",
+                    "reason": "missing",
+                    "format_version": None,
+                    "contexts": [],
+                }
+            epoch = self.get_execution_epoch()
+            contexts = []
+            rows = conn.execute(
+                "SELECT r.*,c.positive_snapshot_id,c.negative_snapshot_id FROM "
+                "execution_run_contexts r JOIN execution_contexts c ON "
+                "c.context_id=r.context_id WHERE r.run_id=? ORDER BY r.anchor_id",
+                (run_id,),
+            )
+            for row in rows:
+                positive = _values(conn, row["positive_snapshot_id"])
+                negative = _values(conn, row["negative_snapshot_id"])
+                hints = (None, None, None)
+                if row["reference_epoch"] == epoch:
+                    current = _hints(
+                        conn,
+                        positive["raw"] if positive else None,
+                        negative["raw"] if negative else None,
+                    )
+                    hints = tuple(
+                        row[key] if row[key] == value else None
+                        for key, value in zip(
+                            ("positive_prompt_id", "negative_prompt_id", "pair_id"),
+                            current,
+                        )
+                    )
+                contexts.append(
+                    {
+                        "anchor_id": row["anchor_id"],
+                        "context_id": row["context_id"],
+                        "positive": positive,
+                        "negative": negative,
+                        "evidence": json.loads(row["evidence_json"]),
+                        "positive_prompt_id": hints[0],
+                        "negative_prompt_id": hints[1],
+                        "pair_id": hints[2],
+                        "reference_epoch": row["reference_epoch"],
+                    }
+                )
+            return {
+                "state": capture[0],
+                "reason": capture[1],
+                "format_version": capture[2],
+                "contexts": contexts,
+            }
+        finally:
+            if own_transaction:
+                conn.rollback()

database/execution_history.py

@@ -0,0 +1,209 @@
+"""Transactional attempt writes and process-local restore invalidation."""
+
+import os
+import sqlite3
+import threading
+from contextlib import contextmanager
+from typing import Callable, Iterator
+
+try:
+    from ..utils.execution_history_events import (
+        RunFinish,
+        RunStart,
+        classify_evidence,
+        encode_command,
+    )
+    from ..utils.logging_config import get_logger
+except ImportError:
+    from utils.execution_history_events import (
+        RunFinish,
+        RunStart,
+        classify_evidence,
+        encode_command,
+    )
+    from utils.logging_config import get_logger
+
+
+_logger = get_logger("prompt_manager.database.execution_history")
+# Bounded wait for a history write lock. History is best effort and must
+# never stall ComfyUI's queue: a write that cannot start in time raises
+# "database is locked" and the writer records a gap instead of waiting.
+EXECUTION_BUSY_TIMEOUT_MS = 250
+_restore_lock = threading.Lock()
+_restore_observers: dict[str, dict[object, Callable[[], None]]] = {}
+
+
+def _database_identity(db_path: str) -> str:
+    return os.path.normcase(os.path.realpath(db_path))
+
+
+def subscribe_history_restore(
+    db_path: str, callback: Callable[[], None]
+) -> Callable[[], None]:
+    """Observe successful restores of the canonical database path."""
+    identity = _database_identity(db_path)
+    token = object()
+    with _restore_lock:
+        _restore_observers.setdefault(identity, {})[token] = callback
+
+    def unsubscribe() -> None:
+        with _restore_lock:
+            observers = _restore_observers.get(identity)
+            if observers is not None:
+                observers.pop(token, None)
+                if not observers:
+                    _restore_observers.pop(identity, None)
+
+    return unsubscribe
+
+
+def notify_history_restore(db_path: str) -> None:
+    """Notify outside the registry lock; observers cannot fail a restore."""
+    with _restore_lock:
+        callbacks = tuple(
+            _restore_observers.get(_database_identity(db_path), {}).values()
+        )
+    for callback in callbacks:
+        try:
+            callback()
+        except Exception:
+            _logger.warning("Execution history restore observer failed", exc_info=True)
+
+
+class HistoryEpochChanged(Exception):
+    """A command belongs to storage replaced by a completed restore."""
+
+    def __init__(self, storage_epoch: str):
+        self.storage_epoch = storage_epoch
+        super().__init__("Execution history storage epoch changed")
+
+
+class ExecutionHistoryOperations:
+    """Mixin using the calling thread's existing managed model connection."""
+
+    def get_execution_epoch(self) -> str:
+        row = (
+            self.model.get_connection()
+            .execute(
+                "SELECT storage_epoch FROM execution_history_metadata "
+                "WHERE singleton_id=1"
+            )
+            .fetchone()
+        )
+        if row is None:
+            raise sqlite3.DatabaseError("Missing execution history metadata")
+        return row[0]
+
+    @contextmanager
+    def _execution_transaction(
+        self, expected_epoch: str
+    ) -> Iterator[sqlite3.Connection]:
+        conn = self.model.get_connection()
+        if conn.in_transaction:
+            raise sqlite3.ProgrammingError(
+                "Execution history requires its own transaction"
+            )
+        timeout = conn.execute("PRAGMA busy_timeout").fetchone()[0]
+        conn.execute(f"PRAGMA busy_timeout={int(EXECUTION_BUSY_TIMEOUT_MS)}")
+        try:
+            conn.execute("BEGIN IMMEDIATE")
+            try:
+                epoch = self.get_execution_epoch()
+                if epoch != expected_epoch:
+                    raise HistoryEpochChanged(epoch)
+                yield conn
+                conn.commit()
+            except BaseException:
+                conn.rollback()
+                raise
+        finally:
+            conn.execute(f"PRAGMA busy_timeout={timeout}")
+
+    def write_execution_start(self, command: RunStart) -> bool:
+        encode_command(command)
+        if type(command) is not RunStart:
+            raise ValueError("Expected execution start command")
+        with self._execution_transaction(command.storage_epoch) as conn:
+            conn.execute(
+                "INSERT INTO execution_runs "
+                "(run_id,session_id,queue_item_id,job_id,storage_epoch,record_version,"
+                "observed_dequeue_at,outcome,diagnostics_omitted) "
+                "VALUES(?,?,?,?,?,1,?,'observed',0) "
+                "ON CONFLICT(session_id,queue_item_id) DO NOTHING",
+                (
+                    command.run_id,
+                    command.session_id,
+                    command.queue_item_id,
+                    command.job_id,
+                    command.storage_epoch,
+                    command.observed_dequeue_at,
+                ),
+            )
+        return True
+
+    def write_execution_finish(self, command: RunFinish) -> bool:
+        encode_command(command)
+        if type(command) is not RunFinish:
+            raise ValueError("Expected execution finish command")
+        with self._execution_transaction(command.storage_epoch) as conn:
+            row = conn.execute(
+                "SELECT * FROM execution_runs WHERE session_id=? AND queue_item_id=? "
+                "AND storage_epoch=?",
+                (command.session_id, command.queue_item_id, command.storage_epoch),
+            ).fetchone()
+            if row is None:
+                return False
+            completion = classify_evidence(row["job_id"], command.evidence)
+            if row["outcome"] == "unknown":
+                return True
+            if row["outcome"] != "observed":
+                contradictory = completion.reason == "conflicting_terminal" or (
+                    completion.outcome in ("succeeded", "failed", "interrupted")
+                    and completion.outcome != row["outcome"]
+                )
+                if contradictory:
+                    conn.execute(
+                        "UPDATE execution_runs SET outcome='unknown',"
+                        "reason='conflicting_terminal' WHERE run_id=?",
+                        (row["run_id"],),
+                    )
+                return True
+            conn.execute(
+                "UPDATE execution_runs SET observed_callback_at=?,"
+                "executor_started_at=?,"
+                "terminal_event_at=?,outcome=?,reason=?,diagnostics_omitted=?,"
+                "node_id=?,"
+                "node_type=?,exception_type=? WHERE run_id=?",
+                (
+                    command.observed_callback_at,
+                    completion.executor_started_at,
+                    completion.terminal_event_at,
+                    completion.outcome,
+                    completion.reason,
+                    int(completion.diagnostics_omitted),
+                    completion.node_id,
+                    completion.node_type,
+                    completion.exception_type,
+                    row["run_id"],
+                ),
+            )
+        return True
+
+    def get_execution_run(self, run_id: str) -> dict | None:
+        row = (
+            self.model.get_connection()
+            .execute("SELECT * FROM execution_runs WHERE run_id=?", (run_id,))
+            .fetchone()
+        )
+        return dict(row) if row is not None else None
+
+
+def rekey_history_restore(old_path: str, new_path: str) -> None:
+    """Move the observers of ``old_path`` to ``new_path`` (Move library)."""
+    old, new = _database_identity(old_path), _database_identity(new_path)
+    if old == new:
+        return
+    with _restore_lock:
+        observers = _restore_observers.pop(old, None)
+        if observers:
+            _restore_observers.setdefault(new, {}).update(observers)

database/execution_history_schema.py

@@ -0,0 +1,349 @@
+"""Versioned additive schema for observed execution attempts."""
+
+import re
+import sqlite3
+from uuid import uuid4
+
+SCHEMA_VERSION = 3
+
+
+def _uuid_check(column: str) -> str:
+    return (
+        f"typeof({column}) = 'text' AND length({column}) = 36 "
+        f"AND substr({column}, 9, 1) = '-' AND substr({column}, 14, 1) = '-' "
+        f"AND substr({column}, 19, 1) = '-' AND substr({column}, 24, 1) = '-' "
+        f"AND length(replace({column}, '-', '')) = 32 "
+        f"AND replace({column}, '-', '') NOT GLOB '*[^0-9a-f]*'"
+    )
+
+
+def _integer_check(column: str, *, nullable: bool = False) -> str:
+    expression = f"typeof({column}) = 'integer' AND {column} >= 0"
+    return f"{column} IS NULL OR ({expression})" if nullable else expression
+
+
+_METADATA_CHECKS = (
+    "singleton_id = 1",
+    "schema_version = 3",
+    _uuid_check("storage_epoch"),
+)
+_RUN_CHECKS = (
+    *(
+        _uuid_check(column)
+        for column in ("run_id", "session_id", "job_id", "storage_epoch")
+    ),
+    _integer_check("queue_item_id"),
+    "record_version = 1",
+    _integer_check("observed_dequeue_at"),
+    *(
+        _integer_check(column, nullable=True)
+        for column in (
+            "observed_callback_at",
+            "executor_started_at",
+            "terminal_event_at",
+        )
+    ),
+    "outcome IN ('observed', 'succeeded', 'failed', 'interrupted', 'unknown')",
+    "(outcome = 'observed' AND reason IS NULL AND "
+    "observed_callback_at IS NULL) OR (outcome != 'observed' AND "
+    "reason IS NOT NULL AND observed_callback_at IS NOT NULL AND "
+    "((outcome = 'succeeded' AND reason = 'terminal_success') OR "
+    "(outcome = 'failed' AND reason = 'terminal_error') OR (outcome = "
+    "'interrupted' AND reason = 'terminal_interrupted') OR (outcome = "
+    "'unknown' AND reason IN ('missing_terminal', "
+    "'conflicting_terminal', 'invalid_evidence', 'evidence_limit'))))",
+    "diagnostics_omitted IN (0, 1)",
+    *(
+        f"{column} IS NULL OR (typeof({column}) = 'text' "
+        f"AND length({column}) <= {limit})"
+        for column, limit in (
+            ("node_id", 128),
+            ("node_type", 128),
+            ("exception_type", 256),
+        )
+    ),
+)
+_TABLE_COLUMNS = {
+    "execution_history_metadata": (
+        ("singleton_id", "INTEGER", 1, 1),
+        ("schema_version", "INTEGER", 1, 0),
+        ("storage_epoch", "TEXT", 1, 0),
+    ),
+    "execution_runs": (
+        ("run_id", "TEXT", 1, 1),
+        ("session_id", "TEXT", 1, 0),
+        ("queue_item_id", "INTEGER", 1, 0),
+        ("job_id", "TEXT", 1, 0),
+        ("storage_epoch", "TEXT", 1, 0),
+        ("record_version", "INTEGER", 1, 0),
+        ("observed_dequeue_at", "INTEGER", 1, 0),
+        ("observed_callback_at", "INTEGER", 0, 0),
+        ("executor_started_at", "INTEGER", 0, 0),
+        ("terminal_event_at", "INTEGER", 0, 0),
+        ("outcome", "TEXT", 1, 0),
+        ("reason", "TEXT", 0, 0),
+        ("diagnostics_omitted", "INTEGER", 1, 0),
+        ("node_id", "TEXT", 0, 0),
+        ("node_type", "TEXT", 0, 0),
+        ("exception_type", "TEXT", 0, 0),
+    ),
+}
+_TABLE_CHECKS = {
+    "execution_history_metadata": _METADATA_CHECKS,
+    "execution_runs": _RUN_CHECKS,
+}
+# Row-only guards preserve the supported v1 DDL fingerprint while bounding reads.
+_ROW_BYTE_CHECKS = {
+    "execution_history_metadata": ("length(CAST(storage_epoch AS BLOB)) = 36",),
+    "execution_runs": (
+        *(
+            f"length(CAST({column} AS BLOB)) = 36"
+            for column in (
+                "run_id",
+                "session_id",
+                "job_id",
+                "storage_epoch",
+            )
+        ),
+        *(
+            f"{column} IS NULL OR length(CAST({column} AS BLOB)) <= {limit}"
+            for column, limit in (
+                ("node_id", 512),
+                ("node_type", 512),
+                ("exception_type", 1024),
+            )
+        ),
+    ),
+}
+
+_INDEXES = {
+    "idx_execution_runs_identity": (True, ("session_id", "queue_item_id")),
+    "idx_execution_runs_observed": (False, ("observed_dequeue_at", "run_id")),
+}
+
+
+def _compact(sql: str) -> str:
+    return "".join(
+        token if token.startswith("'") else re.sub(r"\s+", "", token).lower()
+        for token in re.findall(r"'(?:''|[^'])*'|[^']+", sql)
+    )
+
+
+def history_schema_error(
+    conn: sqlite3.Connection, *, allow_missing: bool = True
+) -> str | None:
+    """Reject partial/unsupported history schemas and invalid persisted records."""
+    tables = {
+        row[0]: row[1]
+        for row in conn.execute("SELECT name,sql FROM sqlite_master WHERE type='table'")
+        if row[0] in _TABLE_COLUMNS
+    }
+    from .execution_context_schema import TABLE_COLUMNS, context_schema_error
+
+    from .execution_output_schema import (
+        TABLE_COLUMNS as OUTPUT_COLUMNS,
+        output_schema_error,
+    )
+
+    output_present = any(
+        row[0] in OUTPUT_COLUMNS
+        for row in conn.execute("SELECT name FROM sqlite_master")
+    )
+    if not tables and output_present:
+        return "Execution outputs require execution history"
+    context_present = any(
+        row[0] in TABLE_COLUMNS
+        for row in conn.execute("SELECT name FROM sqlite_master WHERE type='table'")
+    )
+    if not tables and context_present:
+        return "Execution contexts require execution history"
+    if not tables:
+        return None if allow_missing else "Missing execution history schema"
+    if tables.keys() != _TABLE_COLUMNS.keys():
+        return "Incomplete execution history schema"
+    metadata = conn.execute(
+        "SELECT CASE WHEN typeof(schema_version) = 'integer' "
+        "THEN schema_version ELSE NULL END FROM execution_history_metadata LIMIT 2"
+    ).fetchall()
+    if len(metadata) != 1 or len(metadata[0]) != 1 or metadata[0][0] not in (1, 2, 3):
+        return "Unsupported execution history metadata/version"
+    version = metadata[0][0]
+    if version < 3 and output_present:
+        return "Unexpected execution outputs for history v1/v2"
+    if version == 1 and context_present:
+        return "Unexpected execution contexts for history v1"
+    if not allow_missing and version != SCHEMA_VERSION:
+        return "Execution history migration required"
+    for table, expected in _TABLE_COLUMNS.items():
+        columns = tuple(
+            (row[1], row[2].upper(), row[3], row[5])
+            for row in conn.execute(f"PRAGMA table_info({table})")
+        )
+        if columns != expected:
+            return f"{table} has invalid execution history columns"
+        sql = _compact(tables[table])
+        checks = _TABLE_CHECKS[table]
+        if table == "execution_history_metadata":
+            checks = tuple(
+                check.replace("schema_version = 3", f"schema_version = {version}")
+                for check in checks
+            )
+        if sql.count("check(") != len(checks) or any(
+            f"check({_compact(check)})" not in sql for check in checks
+        ):
+            return f"{table} has invalid execution history checks"
+        if conn.execute(f"PRAGMA foreign_key_list({table})").fetchone():
+            return f"{table} has unexpected execution history foreign keys"
+        for index in conn.execute(f"PRAGMA index_list({table})"):
+            if not index[2]:
+                continue
+            if table == "execution_runs" and index[1] == "idx_execution_runs_identity":
+                continue
+            name = index[1].replace('"', '""')
+            keys = tuple(
+                row for row in conn.execute(f'PRAGMA index_xinfo("{name}")') if row[5]
+            )
+            primary = tuple(column[0] for column in expected if column[3])
+            if (
+                index[3] != "pk"
+                or index[4]
+                or tuple(key[2] for key in keys) != primary
+                or any(key[3] or key[4] != "BINARY" for key in keys)
+            ):
+                return f"{table} has unexpected execution history uniqueness"
+        predicate = " AND ".join(
+            f"({check})" for check in (*checks, *_ROW_BYTE_CHECKS[table])
+        )
+        if conn.execute(
+            f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1"
+        ).fetchone():
+            return f"{table} contains invalid execution history records"
+    metadata = conn.execute(
+        "SELECT singleton_id,schema_version,storage_epoch FROM "
+        "execution_history_metadata"
+    ).fetchall()
+    if len(metadata) != 1 or metadata[0][0] != 1 or metadata[0][1] != version:
+        return "Unsupported execution history metadata/version"
+    indexes = {row[1]: row for row in conn.execute("PRAGMA index_list(execution_runs)")}
+    for name, (unique, columns) in _INDEXES.items():
+        row = indexes.get(name)
+        if row is None or bool(row[2]) != unique or row[4]:
+            return f"Missing or invalid execution history index {name}"
+        actual = tuple(conn.execute(f"PRAGMA index_xinfo({name})"))
+        keys = tuple(item for item in actual if item[5])
+        if tuple(item[2] for item in keys) != columns or any(
+            item[3] or item[4] != "BINARY" for item in keys
+        ):
+            return f"Invalid execution history index {name}"
+    if version >= 2:
+        error = context_schema_error(conn)
+        if error:
+            return error
+    return output_schema_error(conn) if version == 3 else None
+
+
+def install_context_schema(conn: sqlite3.Connection) -> None:
+    from .execution_context_schema import install_context_schema as install
+
+    install(conn)
+
+
+def install_output_schema(conn: sqlite3.Connection) -> None:
+    from .execution_output_schema import install_output_schema as install
+
+    install(conn)
+
+
+def migrate_execution_history(conn: sqlite3.Connection) -> None:
+    """Install history, contexts and outputs atomically, without backfill."""
+    conn.execute("SAVEPOINT execution_history_schema")
+    try:
+        error = history_schema_error(conn)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        present = conn.execute(
+            "SELECT 1 FROM sqlite_master WHERE name IN "
+            "('execution_runs','execution_history_metadata') LIMIT 1"
+        ).fetchone()
+        if not present:
+            for table, columns in _TABLE_COLUMNS.items():
+                definitions = [
+                    f"{name} {kind}{' NOT NULL' if required else ''}"
+                    f"{' PRIMARY KEY' if primary else ''}"
+                    for name, kind, required, primary in columns
+                ]
+                definitions.extend(f"CHECK({check})" for check in _TABLE_CHECKS[table])
+                conn.execute(f"CREATE TABLE {table} ({', '.join(definitions)})")
+            for name, (unique, columns) in _INDEXES.items():
+                conn.execute(
+                    f"CREATE {'UNIQUE ' if unique else ''}INDEX {name} "
+                    f"ON execution_runs({', '.join(columns)})"
+                )
+            conn.execute(
+                "INSERT INTO execution_history_metadata VALUES(1,?,?)",
+                (SCHEMA_VERSION, str(uuid4())),
+            )
+            install_context_schema(conn)
+            install_output_schema(conn)
+        else:
+            version, epoch = conn.execute(
+                "SELECT schema_version,storage_epoch FROM execution_history_metadata"
+            ).fetchone()
+            if version < SCHEMA_VERSION:
+                conn.execute(
+                    "ALTER TABLE execution_history_metadata RENAME TO "
+                    "execution_history_metadata_old"
+                )
+                definitions = [
+                    f"{name} {kind}{' NOT NULL' if required else ''}"
+                    f"{' PRIMARY KEY' if primary else ''}"
+                    for name, kind, required, primary in _TABLE_COLUMNS[
+                        "execution_history_metadata"
+                    ]
+                ]
+                definitions.extend(f"CHECK({check})" for check in _METADATA_CHECKS)
+                conn.execute(
+                    "CREATE TABLE execution_history_metadata "
+                    f"({', '.join(definitions)})"
+                )
+                conn.execute(
+                    "INSERT INTO execution_history_metadata VALUES(1,?,?)",
+                    (SCHEMA_VERSION, epoch),
+                )
+                conn.execute("DROP TABLE execution_history_metadata_old")
+                if version == 1:
+                    install_context_schema(conn)
+                install_output_schema(conn)
+        # Read projections need reverse content lookup and the complete run order.
+        for table, columns in (
+            ("execution_prompt_snapshots", "raw_text_id,snapshot_id"),
+            ("execution_contexts", "positive_snapshot_id,context_id"),
+            ("execution_run_contexts", "context_id,run_id,anchor_id"),
+            ("execution_runs", "observed_dequeue_at,session_id,queue_item_id,run_id"),
+        ):
+            conn.execute(
+                f"CREATE INDEX IF NOT EXISTS idx_history_read_{table} "
+                f"ON {table}({columns})"
+            )
+        error = history_schema_error(conn, allow_missing=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute("RELEASE execution_history_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO execution_history_schema")
+        conn.execute("RELEASE execution_history_schema")
+        raise
+
+
+def rotate_history_epoch(conn: sqlite3.Connection) -> str:
+    """Publish a fresh staging epoch without rewriting restored run epochs."""
+    epoch = str(uuid4())
+    with conn:
+        cursor = conn.execute(
+            "UPDATE execution_history_metadata SET storage_epoch=? WHERE "
+            "singleton_id=1",
+            (epoch,),
+        )
+        if cursor.rowcount != 1:
+            raise sqlite3.DatabaseError("Missing execution history metadata")
+    return epoch

database/execution_output_schema.py

@@ -0,0 +1,218 @@
+"""Owned schema and bounded restore validation for frozen output locators."""
+
+import sqlite3
+
+from .execution_history_schema import _compact, _uuid_check
+
+try:
+    from ..utils.execution_outputs import (
+        BUNDLE_OVERHEAD,
+        MAX_BUNDLE_BYTES,
+        MAX_IMAGE_POSITIONS,
+        MAX_REFERENCE_BYTES,
+        REFERENCE_OVERHEAD,
+        STATE_REASONS,
+        OutputBundle,
+        OutputReference,
+        validate_output_bundle,
+    )
+except ImportError:
+    from utils.execution_outputs import (
+        BUNDLE_OVERHEAD,
+        MAX_BUNDLE_BYTES,
+        MAX_IMAGE_POSITIONS,
+        MAX_REFERENCE_BYTES,
+        REFERENCE_OVERHEAD,
+        STATE_REASONS,
+        OutputBundle,
+        OutputReference,
+        validate_output_bundle,
+    )
+
+
+TABLE_COLUMNS = {
+    "execution_output_capture": (
+        ("run_id", "TEXT", 1, 1),
+        ("state", "TEXT", 1, 0),
+        ("reason", "TEXT", 1, 0),
+        ("format_version", "INTEGER", 1, 0),
+        ("source", "TEXT", 1, 0),
+    ),
+    "execution_run_outputs": (
+        ("run_id", "TEXT", 1, 1),
+        ("output_node_id", "TEXT", 1, 2),
+        ("image_ordinal", "INTEGER", 1, 3),
+        ("folder_type", "TEXT", 1, 0),
+        ("subfolder", "TEXT", 1, 0),
+        ("filename", "TEXT", 1, 0),
+        ("path_style", "TEXT", 1, 0),
+        ("root_key", "TEXT", 0, 0),
+    ),
+}
+TABLE_FOREIGN_KEYS = {
+    "execution_output_capture": (("run_id", "execution_runs", "run_id"),),
+    "execution_run_outputs": (("run_id", "execution_output_capture", "run_id"),),
+}
+
+
+def _text_check(column: str, maximum: int, *, characters: int | None = None) -> str:
+    return (
+        f"typeof({column}) = 'text' AND length({column}) <= "
+        f"{maximum if characters is None else characters} "
+        f"AND length(CAST({column} AS BLOB)) <= {maximum} "
+        f"AND instr(CAST({column} AS BLOB), x'00') = 0"
+    )
+
+
+def _enum_check(column: str, values: tuple | frozenset) -> str:
+    return (
+        f"{_text_check(column, 32)} AND {column} IN ("
+        + ",".join("'" + value + "'" for value in sorted(values))
+        + ")"
+    )
+
+
+_RUN_ID_CHECK = f"{_uuid_check('run_id')} AND {_text_check('run_id', 36)}"
+TABLE_CHECKS = {
+    "execution_output_capture": (
+        _RUN_ID_CHECK,
+        _enum_check("state", tuple(STATE_REASONS)),
+        _enum_check("reason", frozenset.union(*STATE_REASONS.values())),
+        "typeof(format_version) = 'integer' AND format_version = 1",
+        f"{_text_check('source', 18)} AND source = 'host_final_history'",
+    ),
+    "execution_run_outputs": (
+        _RUN_ID_CHECK,
+        f"{_text_check('output_node_id', 512, characters=128)} "
+        "AND length(output_node_id) >= 1",
+        "typeof(image_ordinal) = 'integer' AND image_ordinal BETWEEN 0 AND 255",
+        _enum_check("folder_type", ("output", "temp")),
+        _text_check("subfolder", 2048),
+        f"{_text_check('filename', 1024)} AND length(filename) >= 1",
+        _enum_check("path_style", ("posix", "windows")),
+        f"root_key IS NULL OR ({_text_check('root_key', 64)} "
+        "AND length(root_key) = 64 AND root_key NOT GLOB '*[^0-9a-f]*')",
+    ),
+}
+
+
+def _table_sql(table: str) -> str:
+    columns = TABLE_COLUMNS[table]
+    definitions = [
+        f"{name} {kind}{' NOT NULL' if required else ''}"
+        for name, kind, required, _ in columns
+    ]
+    primary = [name for name, _, _, order in columns if order]
+    definitions.append(f"PRIMARY KEY ({', '.join(primary)})")
+    definitions.extend(f"CHECK({check})" for check in TABLE_CHECKS[table])
+    definitions.extend(
+        f"FOREIGN KEY ({column}) REFERENCES {target}({key}) "
+        "ON UPDATE NO ACTION ON DELETE NO ACTION"
+        for column, target, key in TABLE_FOREIGN_KEYS[table]
+    )
+    return f"CREATE TABLE {table} ({', '.join(definitions)})"
+
+
+def install_output_schema(conn: sqlite3.Connection) -> None:
+    """Install the two owned tables within the caller's migration transaction."""
+    for table in TABLE_COLUMNS:
+        conn.execute(_table_sql(table))
+
+
+def output_schema_error(conn: sqlite3.Connection) -> str | None:
+    """Reject altered schemas and invalid rows before returning bounded DTOs."""
+    if conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type IN ('trigger','view') LIMIT 1"
+    ).fetchone():
+        return "Unsupported execution output trigger/view"
+    tables = {
+        row[0]: row[1]
+        for row in conn.execute("SELECT name,sql FROM sqlite_master WHERE type='table'")
+        if row[0] in TABLE_COLUMNS
+    }
+    if tables.keys() != TABLE_COLUMNS.keys():
+        return "Incomplete execution output schema"
+    for table, expected in TABLE_COLUMNS.items():
+        columns = tuple(
+            (row[1], row[2].upper(), row[3], row[5])
+            for row in conn.execute(f"PRAGMA table_xinfo({table})")
+        )
+        if columns != expected or _compact(tables[table]) != _compact(
+            _table_sql(table)
+        ):
+            return f"{table} has invalid execution output schema"
+        foreign = tuple(
+            sorted(
+                (row[3], row[2], row[4], row[5], row[6], row[7])
+                for row in conn.execute(f"PRAGMA foreign_key_list({table})")
+            )
+        )
+        if foreign != tuple(
+            (*key, "NO ACTION", "NO ACTION", "NONE")
+            for key in TABLE_FOREIGN_KEYS[table]
+        ):
+            return f"{table} has invalid execution output foreign keys"
+        for index in conn.execute(f"PRAGMA index_list({table})"):
+            name = index[1].replace('"', '""')
+            keys = tuple(
+                row for row in conn.execute(f'PRAGMA index_xinfo("{name}")') if row[5]
+            )
+            if (
+                not index[2]
+                or index[3] != "pk"
+                or index[4]
+                or tuple(key[2] for key in keys)
+                != tuple(column[0] for column in expected if column[3])
+                or any(key[3] or key[4] != "BINARY" for key in keys)
+            ):
+                return f"{table} has unexpected execution output index/uniqueness"
+        # Run every scalar guard in SQLite before any evidence text is fetched.
+        predicate = " AND ".join(f"({check})" for check in TABLE_CHECKS[table])
+        if conn.execute(
+            f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1"
+        ).fetchone():
+            return f"{table} contains invalid execution output records"
+        if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
+            return f"{table} contains orphan execution output records"
+    reference_bytes = f"{REFERENCE_OVERHEAD} + " + " + ".join(
+        f"coalesce(length(CAST({column} AS BLOB)), 0)"
+        for column in (
+            "output_node_id",
+            "folder_type",
+            "subfolder",
+            "filename",
+            "path_style",
+            "root_key",
+        )
+    )
+    if conn.execute(
+        "SELECT 1 FROM execution_run_outputs "
+        f"WHERE ({reference_bytes}) > {MAX_REFERENCE_BYTES} LIMIT 1"
+    ).fetchone():
+        return "Oversized execution output reference"
+    if conn.execute(
+        "SELECT 1 FROM execution_run_outputs GROUP BY run_id "
+        f"HAVING COUNT(*) > {MAX_IMAGE_POSITIONS} OR "
+        f"{BUNDLE_OVERHEAD} + SUM({reference_bytes}) > {MAX_BUNDLE_BYTES} LIMIT 1"
+    ).fetchone():
+        return "Oversized execution output bundle"
+    try:
+        for run_id, state, reason, version, source in conn.execute(
+            "SELECT run_id,state,reason,format_version,source "
+            "FROM execution_output_capture"
+        ):
+            references = tuple(
+                OutputReference(*row)
+                for row in conn.execute(
+                    "SELECT output_node_id,image_ordinal,folder_type,subfolder,"
+                    "filename,path_style,root_key FROM execution_run_outputs "
+                    "WHERE run_id=? ORDER BY output_node_id,image_ordinal",
+                    (run_id,),
+                )
+            )
+            validate_output_bundle(
+                OutputBundle(references, state, reason, version, source)
+            )
+    except (ValueError, TypeError, OverflowError, sqlite3.Error) as exc:
+        return f"Invalid execution output content: {type(exc).__name__}"
+    return None

database/execution_outputs.py

@@ -0,0 +1,153 @@
+"""Immutable run-owned output references, independent of mutable galleries."""
+
+import sqlite3
+from contextlib import ExitStack
+from typing import Callable, ContextManager
+
+try:
+    from ..utils.execution_history_events import RunFinish, encode_command
+    from ..utils.execution_outputs import OutputBundle, validate_output_bundle
+except ImportError:
+    from utils.execution_history_events import RunFinish, encode_command
+    from utils.execution_outputs import OutputBundle, validate_output_bundle
+
+
+_REFERENCE_FIELDS = (
+    "output_node_id",
+    "image_ordinal",
+    "folder_type",
+    "subfolder",
+    "filename",
+    "path_style",
+    "root_key",
+)
+_REFERENCE_COLUMNS = ",".join(_REFERENCE_FIELDS)
+
+
+def _capture(conn: sqlite3.Connection, run_id: str) -> sqlite3.Row | None:
+    capture = conn.execute(
+        "SELECT state,reason,format_version,source FROM execution_output_capture "
+        "WHERE run_id=?",
+        (run_id,),
+    ).fetchone()
+    # Legacy callers may disable FK enforcement; never adopt orphan evidence.
+    if capture is None:
+        if conn.execute(
+            "SELECT 1 FROM execution_run_outputs WHERE run_id=? LIMIT 1", (run_id,)
+        ).fetchone():
+            raise sqlite3.DatabaseError("Output references have no capture header")
+    elif (
+        conn.execute(
+            "SELECT 1 FROM execution_runs WHERE run_id=?", (run_id,)
+        ).fetchone()
+        is None
+    ):
+        raise sqlite3.DatabaseError("Output capture has no execution parent")
+    return capture
+
+
+def _references(conn: sqlite3.Connection, run_id: str) -> list[tuple]:
+    return [
+        tuple(row)
+        for row in conn.execute(
+            f"SELECT {_REFERENCE_COLUMNS} FROM execution_run_outputs "
+            "WHERE run_id=? ORDER BY output_node_id,image_ordinal",
+            (run_id,),
+        )
+    ]
+
+
+class ExecutionOutputOperations:
+    """Compose with ExecutionHistoryOperations on PromptDatabase."""
+
+    def write_execution_outputs_for_finish(
+        self,
+        command: RunFinish,
+        bundle: OutputBundle,
+        *,
+        generation_guard: Callable[[], ContextManager[bool]] | None = None,
+    ) -> bool:
+        encode_command(command)
+        if type(command) is not RunFinish:
+            raise ValueError("Expected execution finish command")
+        validate_output_bundle(bundle)
+        header = (bundle.state, bundle.reason, bundle.format_version, bundle.source)
+        prepared = [
+            tuple(getattr(reference, field) for field in _REFERENCE_FIELDS)
+            for reference in bundle.references
+        ]
+        # Enter after BEGIN and retain the guard until COMMIT/ROLLBACK finishes.
+        with ExitStack() as guards:
+            with self._execution_transaction(command.storage_epoch) as conn:
+                if generation_guard is not None and not guards.enter_context(
+                    generation_guard()
+                ):
+                    return False
+                parent = conn.execute(
+                    "SELECT run_id FROM execution_runs WHERE session_id=? "
+                    "AND queue_item_id=? AND storage_epoch=?",
+                    (command.session_id, command.queue_item_id, command.storage_epoch),
+                ).fetchone()
+                if parent is None:
+                    return False
+                run_id = parent[0]
+                capture = _capture(conn, run_id)
+                if capture is not None:
+                    if capture[0] != "conflicting" and (
+                        tuple(capture) != header
+                        or _references(conn, run_id) != prepared
+                    ):
+                        conn.execute(
+                            "UPDATE execution_output_capture SET "
+                            "state='conflicting',reason='conflicting' WHERE run_id=?",
+                            (run_id,),
+                        )
+                    return True
+                conn.execute(
+                    "INSERT INTO execution_output_capture "
+                    "(run_id,state,reason,format_version,source) VALUES(?,?,?,?,?)",
+                    (run_id, *header),
+                )
+                conn.executemany(
+                    f"INSERT INTO execution_run_outputs (run_id,{_REFERENCE_COLUMNS}) "
+                    "VALUES(?,?,?,?,?,?,?,?)",
+                    [(run_id, *reference) for reference in prepared],
+                )
+        return True
+
+    def get_execution_outputs(self, run_id: str) -> dict:
+        """Read frozen evidence in one snapshot without resolving its locators."""
+        conn = self.model.get_connection()
+        own_transaction = not conn.in_transaction
+        if own_transaction:
+            conn.execute("BEGIN")
+        try:
+            capture = _capture(conn, run_id)
+            if capture is None:
+                return {
+                    "state": "not_captured",
+                    "reason": None,
+                    "format_version": None,
+                    "source": None,
+                    "references": [],
+                }
+            references = []
+            for row in _references(conn, run_id):
+                reference = dict(zip(_REFERENCE_FIELDS, row))
+                reference.update(
+                    file_identity="unverified",
+                    newly_generated="unknown",
+                    context_membership="unknown",
+                    transient=reference["folder_type"] == "temp",
+                )
+                references.append(reference)
+            return {
+                "state": capture[0],
+                "reason": capture[1],
+                "format_version": capture[2],
+                "source": capture[3],
+                "references": references,
+            }
+        finally:
+            if own_transaction:
+                conn.rollback()

database/health.py

@@ -0,0 +1,380 @@
+"""The read-only health check behind System › Overview and Maintenance.
+
+``run_health_checks`` opens its own ``mode=ro`` connection, reads one
+snapshot and closes the connection in ``finally``; it never writes. Each
+check reports ``passed``, ``issues`` (with an exact ``count``; the
+integrity check stops at ``INTEGRITY_LIMIT`` problems and then says
+``capped``),
+``unavailable`` (it could not be checked here, so ``count`` is None, or
+``missing_files`` checked only some sources and found nothing) or ``failed``
+(it raised; the other checks still run). ``missing_files`` also reports each
+source and whether the check was ``partial``; it reads the image rows inside
+the snapshot and checks their files after the snapshot ended, so a slow source
+never holds the WAL. At most ``MAX_DETAIL`` detail strings are returned per
+check, and none of them carries a server path.
+"""
+
+import datetime
+import logging
+import ntpath
+import os
+import sqlite3
+from pathlib import PurePosixPath, PureWindowsPath
+from urllib.parse import quote
+
+try:
+    from ..utils.hashing import generate_prompt_hash
+    from .models import normalize_image_path
+except ImportError:
+    from utils.hashing import generate_prompt_hash
+    from database.models import normalize_image_path
+
+logger = logging.getLogger("prompt_manager.database.health")
+
+MAX_DETAIL = 10  # detail strings returned per check
+INTEGRITY_LIMIT = 100  # PRAGMA integrity_check(N): problems SQLite reports
+BATCH_ROWS = 100  # image rows read per fetch
+REVIEW_UNAVAILABLE = "Needs review is not available yet."
+CHECK_FAILED = "The check could not finish."
+# Check ids and the labels of the concept's health report, in its order;
+# pair uniqueness and consistency are the added rows (D63, now D95).
+CHECKS = (
+    ("integrity", "Integrity check"),
+    ("foreign_keys", "Foreign keys & pair references"),
+    ("canonical_hashes", "Canonical hashes"),
+    ("pair_uniqueness", "Pair uniqueness"),
+    ("missing_files", "Missing files"),
+    ("ambiguous_metadata", "Ambiguous metadata"),
+    ("consistency", "Consistency"),
+)
+_DATABASE_CHECKS = (
+    "integrity",
+    "foreign_keys",
+    "canonical_hashes",
+    "pair_uniqueness",
+    "missing_files",
+    "consistency",
+)
+
+
+def iter_consistency_issues(conn):
+    """The legacy consistency issues: tag and image rows of a missing prompt."""
+    for ref in conn.execute(
+        "SELECT pt.prompt_id FROM prompt_tags pt "
+        "LEFT JOIN prompts p ON pt.prompt_id = p.id WHERE p.id IS NULL"
+    ):
+        yield f"prompt_tags entry references non-existent prompt {ref[0]}"
+    for ref in conn.execute(
+        "SELECT gi.id, gi.prompt_id FROM generated_images gi "
+        "LEFT JOIN prompts p ON gi.prompt_id = p.id WHERE p.id IS NULL"
+    ):
+        yield f"Image {ref[0]} references non-existent prompt {ref[1]}"
+
+
+def _result(count, detail, *, status=None):
+    return {
+        "status": status or ("issues" if count else "passed"),
+        "count": count,
+        "detail": detail[:MAX_DETAIL],
+    }
+
+
+def _counted(lines):
+    """(count, first MAX_DETAIL lines) of an iterable, consumed once."""
+    count, detail = 0, []
+    for line in lines:
+        count += 1
+        if len(detail) < MAX_DETAIL:
+            detail.append(line)
+    return _result(count, detail)
+
+
+def readonly_uri(db_path, *, windows=None):
+    """The ``mode=ro`` SQLite URI of *db_path*, on this platform or the given one.
+
+    pathlib spells a UNC path ``\\\\server\\share\\x.db`` as
+    ``file://server/share/x.db``, a remote authority SQLite refuses to open.
+    With an empty authority (``file:////server/share/x.db``) SQLite hands
+    Windows the ``//server/share`` path, which it opens as the UNC path.
+    """
+    if windows is None:
+        windows = os.name == "nt"
+        db_path = os.path.abspath(db_path)
+    if not windows:
+        return PurePosixPath(db_path).as_uri() + "?mode=ro"
+    path = PureWindowsPath(db_path)
+    if ntpath.splitdrive(str(path))[0].startswith("\\\\"):
+        return "file:" + quote("//" + path.as_posix()) + "?mode=ro"
+    return path.as_uri() + "?mode=ro"
+
+
+def _integrity(conn, _roots):
+    rows = [
+        row[0] for row in conn.execute(f"PRAGMA integrity_check({INTEGRITY_LIMIT})")
+    ]
+    if rows == ["ok"]:
+        return {**_result(0, []), "capped": False}
+    # SQLite stops listing at the limit: there may be more problems than that.
+    return {**_result(len(rows), rows), "capped": len(rows) >= INTEGRITY_LIMIT}
+
+
+def _foreign_keys(conn, _roots):
+    return _counted(
+        f"{table} row {rowid} references a missing {parent} row"
+        for table, rowid, parent, _fkid in conn.execute("PRAGMA foreign_key_check")
+    )
+
+
+def _prompt_hash(text):
+    return generate_prompt_hash(text) if isinstance(text, str) else None
+
+
+def _canonical_hashes(conn, _roots):
+    conn.create_function("pm_prompt_hash", 1, _prompt_hash, deterministic=True)
+    return _counted(
+        f"Prompt {row[0]}: the stored hash does not match its text"
+        for row in conn.execute(
+            "SELECT id FROM prompts WHERE hash IS NOT pm_prompt_hash(text) ORDER BY id"
+        )
+    )
+
+
+def _pair_uniqueness(conn, _roots):
+    groups = conn.execute(
+        "SELECT positive_prompt_id,negative_prompt_id,COUNT(*) FROM prompt_pairs "
+        "GROUP BY positive_prompt_id,negative_prompt_id HAVING COUNT(*)>1 "
+        "ORDER BY positive_prompt_id,negative_prompt_id"
+    ).fetchall()
+    detail = [
+        f"Positive prompt {positive} with "
+        + ("no negative" if negative is None else f"negative {negative}")
+        + f": {rows} pair rows"
+        for positive, negative, rows in groups
+    ]
+    # The surplus rows: what keeping one row per pair would leave out.
+    return _result(sum(rows - 1 for _p, _n, rows in groups), detail)
+
+
+def _consistency(conn, _roots):
+    return _counted(iter_consistency_issues(conn))
+
+
+def _readable(root):
+    return os.path.isdir(root) and os.access(root, os.R_OK | os.X_OK)
+
+
+def _root_prefix(root):
+    return normalize_image_path(root).rstrip("\\/") + os.sep
+
+
+def _source_name(root, index):
+    """A source's folder name for the report (never its server path)."""
+    return os.path.basename(str(root).rstrip("\\/")) or f"Source {index + 1}"
+
+
+def _owner(path, prefixes):
+    """Index of the longest source prefix *path* lies under, or None."""
+    owner, longest = None, -1
+    for index, prefix in enumerate(prefixes):
+        if path.startswith(prefix) and len(prefix) > longest:
+            owner, longest = index, len(prefix)
+    return owner
+
+
+def _image_owner(image_path, file_path, prefixes):
+    spelled = file_path or normalize_image_path(image_path)
+    owner = _owner(spelled, prefixes)
+    if owner is None:
+        owner = _owner(normalize_image_path(os.path.realpath(image_path)), prefixes)
+    return owner
+
+
+def readable_roots(roots):
+    """Whether each source can be read now (an unmounted drive cannot)."""
+    return [_readable(root) for root in roots]
+
+
+def link_states(rows, roots, readable):
+    """``(row, owner, state)`` for each image row ``(id, image_path, filename,
+    file_path)``: ``owner`` is the index of the source it lies under (None
+    outside every source) and ``state`` is ``missing``, ``present``,
+    ``unavailable`` (its source cannot be read, so the file is not looked at)
+    or ``outside``. ``readable`` is ``readable_roots(roots)``. The health
+    check's ``missing_files`` and the maintenance missing-link plan both
+    judge links this way, so their counts agree.
+    """
+    prefixes = [_root_prefix(root) for root in roots]
+    for row in rows:
+        owner = _image_owner(row[1], row[3], prefixes)
+        if owner is None:
+            yield row, None, "outside"
+        elif not readable[owner]:
+            yield row, owner, "unavailable"
+        elif os.path.exists(row[1]):
+            yield row, owner, "present"
+        else:
+            yield row, owner, "missing"
+
+
+def _unavailable_line(unavailable, *, partial):
+    noun = "source is" if unavailable == 1 else "sources are"
+    effect = "its files were" if unavailable == 1 else "their files were"
+    if not partial:
+        effect = "missing files were"
+    return f"{unavailable} image {noun} unavailable, so {effect} not counted."
+
+
+def _missing_files(conn, roots):
+    """Links whose file is gone, counted per source (only readable ones).
+
+    Each source reports ``checked`` with its own count, or ``unavailable``
+    (an unmounted drive) with no count: its links are never turned into a
+    count, while the readable sources still are. The check's ``count`` is the
+    readable sources' total (None when none is readable) and ``partial`` says
+    some sources were not checked. Links outside every source are named in the
+    detail, not counted.
+
+    Inside the snapshot this only reads the image rows; it returns the step
+    that checks their files, which runs after the snapshot ended.
+    """
+    if not roots:
+        result = _result(None, ["No image source is available."], status="unavailable")
+        return {**result, "partial": False, "sources": []}
+    rows = []
+    cursor = conn.execute(
+        "SELECT id,image_path,filename,file_path FROM generated_images ORDER BY id"
+    )
+    while batch := cursor.fetchmany(BATCH_ROWS):
+        rows.extend(batch)
+    return lambda: _check_files(rows, roots)
+
+
+def _check_files(rows, roots):
+    """The ``missing_files`` result of the image *rows*, from the file system."""
+    readable = readable_roots(roots)
+    sources = [
+        {
+            "name": _source_name(root, index),
+            "status": "checked" if ok else "unavailable",
+            "count": 0 if ok else None,
+        }
+        for index, (root, ok) in enumerate(zip(roots, readable))
+    ]
+    unavailable = readable.count(False)
+    if not any(readable):
+        result = _result(
+            None, [_unavailable_line(unavailable, partial=False)], status="unavailable"
+        )
+        return {**result, "partial": False, "sources": sources}
+    outside, missing = 0, []
+    for row, owner, state in link_states(rows, roots, readable):
+        if state == "outside":
+            outside += 1
+        elif state == "missing":
+            image_id, image_path, filename, _file_path = row
+            sources[owner]["count"] += 1
+            if len(missing) < MAX_DETAIL:
+                name = filename or os.path.basename(image_path)
+                missing.append(f"Image {image_id}: {name} is missing")
+    count = sum(source["count"] or 0 for source in sources)
+    detail = [_unavailable_line(unavailable, partial=True)] if unavailable else []
+    detail += missing
+    if outside:
+        links = "link" if outside == 1 else "links"
+        verb = "was" if outside == 1 else "were"
+        detail.append(
+            f"{outside} {links} outside the image sources {verb} not checked."
+        )
+    status = None if count or not unavailable else "unavailable"
+    result = _result(count, detail, status=status)
+    return {**result, "partial": bool(unavailable), "sources": sources}
+
+
+def _ambiguous(review_count):
+    if review_count is None:
+        return _result(None, [REVIEW_UNAVAILABLE], status="unavailable")
+    return _result(review_count, [])
+
+
+def overall_status(checks) -> str:
+    """``passed`` only when every check passed; any check not run is incomplete."""
+    statuses = {check["status"] for check in checks}
+    if statuses & {"unavailable", "failed"} or any(
+        check.get("partial") for check in checks
+    ):
+        return "incomplete"
+    return "issues" if "issues" in statuses else "passed"
+
+
+def _failed():
+    return _result(None, [CHECK_FAILED], status="failed")
+
+
+def _run_check(check_id, step, *args):
+    """One check's result, or None (reported as failed) when it raised."""
+    try:
+        return step(*args)
+    except Exception:  # any error fails this check only; the others still run
+        logger.exception(f"Health check {check_id} failed")
+        return None
+
+
+def _run_database_checks(db_path, roots):
+    """Each database check's result (None when it raised), from one snapshot.
+
+    A check may return a step instead of its result: the step runs after the
+    snapshot ended and the connection closed (``missing_files``' file checks).
+    """
+    results = dict.fromkeys(_DATABASE_CHECKS)
+    try:
+        conn = sqlite3.connect(readonly_uri(db_path), uri=True)
+    except sqlite3.Error:
+        logger.exception("Health check could not open the database")
+        return results
+    try:
+        conn.execute("BEGIN")  # one read snapshot for every check
+        for check_id in _DATABASE_CHECKS:
+            runner = globals()["_" + check_id]  # looked up per run (tests patch it)
+            results[check_id] = _run_check(check_id, runner, conn, roots)
+    except sqlite3.Error:
+        logger.exception("Health check could not read the database")
+    finally:
+        conn.close()
+    for check_id, result in results.items():
+        if callable(result):
+            results[check_id] = _run_check(check_id, result)
+    return results
+
+
+class HealthOperations:
+    """Mixin for PromptDatabase: the read-only System health check."""
+
+    def run_health_checks(self, roots, *, review_count=None) -> dict:
+        """Run every check and return a ``HealthReport``.
+
+        Args:
+            roots: The image source directories to judge missing files under:
+                the effective sources plus any configured source that is not
+                available (reported as an unavailable source of ``missing_files``).
+            review_count: The Needs-review pending count, or None while no
+                review queue exists (``ambiguous_metadata`` is then unavailable).
+        """
+        if review_count is not None and (
+            type(review_count) is not int or review_count < 0
+        ):
+            raise ValueError("review_count must be a non-negative integer or None")
+        roots = [str(root) for root in roots]
+        database = _run_database_checks(self.model.db_path, roots)
+        checks = []
+        for check_id, label in CHECKS:
+            if check_id == "ambiguous_metadata":
+                result = _ambiguous(review_count)
+            else:
+                result = database[check_id] or _failed()
+            checks.append({"id": check_id, "label": label, **result})
+        return {
+            "success": True,
+            "checked_at": datetime.datetime.now(datetime.timezone.utc).isoformat(),
+            "status": overall_status(checks),
+            "checks": checks,
+        }

database/history_gallery.py

@@ -0,0 +1,140 @@
+"""Bounded exact-path saved display within the caller's history snapshot."""
+
+import os
+import sqlite3
+
+try:
+    from ..utils.inspection_response import (
+        MAX_LIBRARY_BYTES,
+        _library_projection,
+        library_metadata,
+    )
+except ImportError:
+    from utils.inspection_response import (
+        MAX_LIBRARY_BYTES,
+        _library_projection,
+        library_metadata,
+    )
+
+
+def _display_bytes(value: bytes | None) -> str | None:
+    if value is None:
+        return None
+    return value.decode("utf-8", "replace")
+
+
+def _tags(
+    conn: sqlite3.Connection, owner: int, *, pair: bool
+) -> tuple[list[str], bool]:
+    table, column = (
+        ("prompt_pair_tags", "pair_id") if pair else ("prompt_tags", "prompt_id")
+    )
+    tags = []
+    used = 0
+    cursor = conn.execute(
+        f"SELECT substr(CAST(t.name AS BLOB),1,?) FROM {table} pt "
+        f"JOIN tags t ON t.id=pt.tag_id WHERE pt.{column}=? LIMIT 129",
+        (MAX_LIBRARY_BYTES + 1, owner),
+    )
+    for row in cursor:
+        used += len(row[0])
+        if used > MAX_LIBRARY_BYTES or len(tags) == 128:
+            return [], True
+        tags.append(_display_bytes(row[0]))
+    return tags, False
+
+
+def _saved_projection(conn: sqlite3.Connection, row: sqlite3.Row) -> tuple[dict, dict]:
+    saved = {"prompt_id": row["prompt_id"]}
+    pairs = list(
+        conn.execute(
+            "SELECT pp.id,pp.positive_prompt_id,"
+            "substr(CAST(pp.title AS BLOB),1,?) AS title,"
+            "substr(CAST(pp.category AS BLOB),1,?) AS category "
+            "FROM image_prompt_pairs ip JOIN prompt_pairs pp ON pp.id=ip.pair_id "
+            "WHERE ip.image_id=? LIMIT 2",
+            (MAX_LIBRARY_BYTES + 1, MAX_LIBRARY_BYTES + 1, row["id"]),
+        )
+    )
+    if len(pairs) > 1:
+        return saved, _library_projection({"status": "limited"})
+    pair = pairs[0] if pairs else None
+    title = pair["title"] if pair is not None else None
+    recipe_tags, limited = (
+        _tags(conn, pair["id"], pair=True) if pair is not None else ([], False)
+    )
+    positive_tags = []
+    if not recipe_tags and not limited:
+        # D39: only read the positive's tags when they are what the panel shows.
+        positive_tags, limited = _tags(conn, row["prompt_id"], pair=False)
+    metadata = library_metadata(
+        row["category"],
+        positive_tags,
+        (pair["category"], recipe_tags) if pair is not None else None,
+    )
+    limited = limited or any(
+        value is not None and len(value) > MAX_LIBRARY_BYTES
+        for value in (title, metadata["category"])
+    )
+    if pair is not None:
+        saved["pair"] = {
+            "id": pair["id"],
+            "positive_prompt_id": pair["positive_prompt_id"],
+        }
+    library = _library_projection(
+        {
+            "status": "linked",
+            "positive_id": row["prompt_id"],
+            "pair_id": pair["id"] if pair is not None else None,
+            "title": _display_bytes(title),
+            **metadata,
+            "category": _display_bytes(metadata["category"]),
+        }
+    )
+    if limited:
+        library = _library_projection({"status": "limited", "tags": []})
+    return saved, library
+
+
+def read_history_gallery(
+    conn: sqlite3.Connection, paths: tuple[str, ...]
+) -> dict[str, dict]:
+    """Read at most two rows/path and two pair associations/unique row; no writes."""
+    if not conn.in_transaction:
+        raise sqlite3.ProgrammingError("Saved history display requires caller snapshot")
+    if (
+        type(paths) is not tuple
+        or len(paths) > 100
+        or any(
+            type(path) is not str
+            or not os.path.isabs(path)
+            or path != os.path.normcase(os.path.normpath(path))
+            or "\0" in path
+            for path in paths
+        )
+    ):
+        raise ValueError("Invalid history gallery paths")
+    result = {}
+    for path in dict.fromkeys(paths):
+        rows = list(
+            conn.execute(
+                "SELECT gi.id,gi.prompt_id,"
+                "substr(CAST(p.category AS BLOB),1,?) AS category "
+                "FROM generated_images gi JOIN prompts p ON p.id=gi.prompt_id "
+                "WHERE gi.file_path=? OR gi.image_path=? LIMIT 2",
+                (MAX_LIBRARY_BYTES + 1, path, path),
+            )
+        )
+        match = "none" if not rows else "unique" if len(rows) == 1 else "ambiguous"
+        saved, library = None, _library_projection(
+            {"status": "none" if not rows else "ambiguous"}
+        )
+        if match == "unique":
+            saved, library = _saved_projection(conn, rows[0])
+        result[path] = {
+            "gallery_match": match,
+            "gallery_image_id": rows[0]["id"] if match == "unique" else None,
+            "saved": saved,
+            "library": library,
+        }
+    return result

database/history_reads.py

@@ -0,0 +1,585 @@
+"""Snapshot-owned, bounded projections of immutable execution evidence."""
+
+import codecs
+from contextlib import contextmanager
+from contextvars import ContextVar
+from dataclasses import dataclass
+import hashlib
+import json
+import re
+import sqlite3
+from typing import Iterator
+
+from .execution_contexts import _hints, _values, content_digest
+from .execution_history import HistoryEpochChanged
+from .execution_outputs import _capture, _REFERENCE_COLUMNS
+from .library import _ELIGIBLE
+
+RunKey = tuple[int, str, int, str]
+
+
+class HistoryCaptureChanged(ValueError):
+    """A continuation no longer describes the stored output capture header."""
+
+
+@dataclass(frozen=True)
+class WindowPosition:
+    raw_text_id: str
+    upper: RunKey
+    before: RunKey
+
+
+@dataclass(frozen=True)
+class OutputPosition:
+    capture_fingerprint: str
+    after: tuple[str, int]
+
+
+_DIGEST = re.compile(r"[0-9a-f]{64}")
+_UUID = re.compile(r"[0-9a-f]{8}(?:-[0-9a-f]{4}){3}-[0-9a-f]{12}")
+
+
+def _valid_digest(value: object) -> bool:
+    return type(value) is str and _DIGEST.fullmatch(value) is not None
+
+
+def _valid_run_key(value: object) -> bool:
+    return (
+        type(value) is tuple
+        and len(value) == 4
+        and all(type(value[i]) is int and 0 <= value[i] <= 2**63 - 1 for i in (0, 2))
+        and all(
+            type(value[i]) is str and _UUID.fullmatch(value[i]) is not None
+            for i in (1, 3)
+        )
+    )
+
+
+def _valid_coordinate(value: object) -> bool:
+    if type(value) is not tuple or len(value) != 2:
+        return False
+    node, ordinal = value
+    if not (
+        type(node) is str
+        and 1 <= len(node) <= 128
+        and "\0" not in node
+        and type(ordinal) is int
+        and 0 <= ordinal <= 255
+    ):
+        return False
+    try:
+        return len(node.encode("utf-8")) <= 512
+    except UnicodeError:
+        return False
+
+
+def _validate_window_position(position: WindowPosition | None) -> None:
+    if position is not None and not (
+        type(position) is WindowPosition
+        and _valid_digest(position.raw_text_id)
+        and _valid_run_key(position.upper)
+        and _valid_run_key(position.before)
+    ):
+        raise ValueError("Invalid history window position")
+
+
+def _validate_output_position(position: OutputPosition | None) -> None:
+    if position is not None and not (
+        type(position) is OutputPosition
+        and _valid_digest(position.capture_fingerprint)
+        and _valid_coordinate(position.after)
+    ):
+        raise ValueError("Invalid history output position")
+
+
+_snapshots: ContextVar[tuple] = ContextVar("history_read_snapshots", default=())
+_KEY = "(r.observed_dequeue_at,r.session_id,r.queue_item_id,r.run_id)"
+_ORDER = (
+    "r.observed_dequeue_at DESC,r.session_id COLLATE BINARY DESC,"
+    "r.queue_item_id DESC,r.run_id COLLATE BINARY DESC"
+)
+# Stale-hint rows a newest-pair lookup inspects before it reports none.
+_PAIR_RUN_CANDIDATES = 16
+_MATCHING = (
+    "SELECT DISTINCT rc.run_id FROM execution_prompt_snapshots s "
+    "JOIN execution_contexts c ON c.positive_snapshot_id=s.snapshot_id "
+    "JOIN execution_run_contexts rc ON rc.context_id=c.context_id "
+    "WHERE s.raw_text_id=?"
+)
+
+
+def _limit(limit: int) -> None:
+    if type(limit) is not int or not 1 <= limit <= 100:
+        raise ValueError("History limit must be an integer from 1 to 100")
+
+
+def _key(row: dict) -> RunKey:
+    return tuple(
+        row[name]
+        for name in ("observed_dequeue_at", "session_id", "queue_item_id", "run_id")
+    )
+
+
+def _context_capture(conn: sqlite3.Connection, run_id: str) -> dict:
+    row = conn.execute(
+        "SELECT state,reason,format_version FROM execution_context_capture "
+        "WHERE run_id=?",
+        (run_id,),
+    ).fetchone()
+    return (
+        dict(row)
+        if row is not None
+        else {"state": "not_captured", "reason": "missing", "format_version": None}
+    )
+
+
+def _output_header(conn: sqlite3.Connection, run_id: str) -> tuple[dict, int | None]:
+    row = _capture(conn, run_id)
+    if row is None:
+        return (
+            dict(state="not_captured", reason=None, format_version=None, source=None),
+            None,
+        )
+    count = conn.execute(
+        "SELECT COUNT(*) FROM execution_run_outputs WHERE run_id=?", (run_id,)
+    ).fetchone()[0]
+    return dict(row), count
+
+
+def _list_status(capture: dict) -> str:
+    if capture["state"] == "not_captured":
+        return "not_captured"
+    return "complete" if capture["state"] == "complete" else "incomplete"
+
+
+def _output_summary(conn: sqlite3.Connection, run_id: str) -> dict:
+    capture, count = _output_header(conn, run_id)
+    return dict(
+        capture, retained_occurrence_count=count, list_status=_list_status(capture)
+    )
+
+
+def _fingerprint(capture: dict, count: int | None) -> str:
+    return hashlib.sha256(
+        json.dumps(
+            [
+                1,
+                *(capture[k] for k in ("state", "reason", "format_version", "source")),
+                count,
+            ],
+            ensure_ascii=False,
+            allow_nan=False,
+            separators=(",", ":"),
+        ).encode("utf-8")
+    ).hexdigest()
+
+
+def _receipt(run_id: str, row: sqlite3.Row) -> dict:
+    return {
+        "occurrence": {
+            "run_id": run_id,
+            "output_node_id": row["output_node_id"],
+            "image_ordinal": row["image_ordinal"],
+        },
+        "reference": {
+            key: row[key]
+            for key in (
+                "folder_type",
+                "subfolder",
+                "filename",
+                "path_style",
+                "root_key",
+            )
+        },
+        "file_identity": "unverified",
+        "newly_generated": "unknown",
+        "context_membership": "unknown",
+        "transient": row["folder_type"] == "temp",
+    }
+
+
+class HistoryReadOperations:
+    """Read evidence only inside an explicitly owned, epoch-first snapshot."""
+
+    @contextmanager
+    def recorded_history_snapshot(
+        self, expected_epoch: str | None = None
+    ) -> Iterator[str]:
+        conn = self.model.get_connection()
+        if conn.in_transaction:
+            raise sqlite3.ProgrammingError(
+                "History reads require their own transaction"
+            )
+        conn.execute("BEGIN")
+        token = _snapshots.set((*_snapshots.get(), conn))
+        try:
+            epoch = self.get_execution_epoch()
+            if expected_epoch is not None and epoch != expected_epoch:
+                raise HistoryEpochChanged(epoch)
+            yield epoch
+        finally:
+            try:
+                conn.rollback()
+            finally:
+                _snapshots.reset(token)
+
+    def _history_connection(self) -> sqlite3.Connection:
+        conn = self.model.get_connection()
+        if not conn.in_transaction or conn not in _snapshots.get():
+            raise sqlite3.ProgrammingError(
+                "History read requires recorded_history_snapshot"
+            )
+        return conn
+
+    def list_history_window(
+        self,
+        positive_id: int,
+        *,
+        limit: int = 40,
+        after: WindowPosition | None = None,
+        restart: bool = False,
+    ) -> dict | None:
+        conn = self._history_connection()
+        _limit(limit)
+        _validate_window_position(after)
+        if type(restart) is not bool or (
+            restart and (after is None or after.before != after.upper)
+        ):
+            raise ValueError("Invalid history restart position")
+        saved = conn.execute(
+            "SELECT CASE WHEN typeof(p.text)='text' AND "
+            "length(CAST(p.text AS BLOB))<=65536 THEN p.text ELSE NULL END AS text "
+            f"FROM prompts p WHERE p.id=? AND ({_ELIGIBLE})",
+            (positive_id,),
+        ).fetchone()
+        if saved is None:
+            return None
+        text = saved["text"]
+        raw_id = None if text is None else content_digest("text", (text,))
+        if after is not None and after.raw_text_id != raw_id:
+            raise ValueError("History position text scope changed")
+        group = {
+            "positive_id": positive_id,
+            "raw_text_id": raw_id,
+            "relation": "exact_frozen_raw_text",
+            "scope_status": "uncapturable_text" if text is None else "no_matches",
+        }
+        result = dict(
+            group=group,
+            recorded_run_count=0,
+            items=[],
+            context_summaries=[],
+            has_more=False,
+            next_position=None,
+            resume_position=None,
+        )
+        if (
+            raw_id is None
+            or conn.execute(
+                "SELECT 1 FROM execution_context_texts WHERE text_id=? "
+                "AND text_value=? COLLATE BINARY",
+                (raw_id, text),
+            ).fetchone()
+            is None
+        ):
+            return result
+        where = f"r.run_id IN ({_MATCHING})"
+        params = [raw_id]
+        if after is not None:
+            where += f" AND {_KEY}<=(?,?,?,?)"
+            params.extend(after.upper)
+        count = conn.execute(
+            f"SELECT COUNT(*) FROM execution_runs r WHERE {where}", params
+        ).fetchone()[0]
+        result["recorded_run_count"] = count
+        group["scope_status"] = "matched" if count else "no_matches"
+        if after is not None and not restart:
+            where += f" AND {_KEY}<(?,?,?,?)"
+            params.extend(after.before)
+        rows = [
+            dict(row)
+            for row in conn.execute(
+                f"SELECT r.* FROM execution_runs r WHERE {where} "
+                f"ORDER BY {_ORDER} LIMIT ?",
+                (*params, limit + 1),
+            )
+        ]
+        more = len(rows) > limit
+        rows = rows[:limit]
+        if not rows:
+            return result
+        upper = after.upper if after is not None else _key(rows[0])
+        result["resume_position"] = WindowPosition(raw_id, upper, upper)
+        if more:
+            result["next_position"] = WindowPosition(raw_id, upper, _key(rows[-1]))
+        result["has_more"] = more
+        rows.reverse()
+        by_run = {row["run_id"]: row for row in rows}
+        for row in rows:
+            row["diagnostics_available"] = any(
+                row[field] is not None
+                for field in ("node_id", "node_type", "exception_type")
+            )
+            for field in ("node_id", "node_type", "exception_type"):
+                del row[field]
+            row["diagnostics_omitted"] = bool(row["diagnostics_omitted"])
+            row["context_capture"] = _context_capture(conn, row["run_id"])
+            row["output_summary"] = _output_summary(conn, row["run_id"])
+            row["matching_anchors"] = []
+        placeholders = ",".join("?" for _ in rows)
+        anchors = conn.execute(
+            "SELECT rc.run_id,rc.anchor_id,rc.context_id,rc.evidence_json,"
+            "c.positive_snapshot_id,c.negative_snapshot_id,"
+            "substr(CAST(t.text_value AS BLOB),1,512) AS prefix,"
+            "length(CAST(t.text_value AS BLOB)) AS byte_length "
+            "FROM execution_run_contexts rc JOIN execution_contexts c "
+            "ON c.context_id=rc.context_id JOIN execution_prompt_snapshots p "
+            "ON p.snapshot_id=c.positive_snapshot_id "
+            "LEFT JOIN execution_prompt_snapshots n "
+            "ON n.snapshot_id=c.negative_snapshot_id "
+            "LEFT JOIN execution_context_texts t ON t.text_id=n.raw_text_id "
+            f"WHERE rc.run_id IN ({placeholders}) AND p.raw_text_id=? "
+            "ORDER BY rc.anchor_id COLLATE BINARY,rc.run_id COLLATE BINARY",
+            (*by_run, raw_id),
+        )
+        summaries = {}
+        evidence_by_context = {}
+        for anchor in anchors:
+            reference = {
+                key: anchor[key]
+                for key in (
+                    "anchor_id",
+                    "context_id",
+                    "positive_snapshot_id",
+                    "negative_snapshot_id",
+                )
+            }
+            by_run[anchor["run_id"]]["matching_anchors"].append(reference)
+            context_id = anchor["context_id"]
+            if context_id not in summaries:
+                prefix = b"" if anchor["byte_length"] == 0 else anchor["prefix"]
+                preview = (
+                    None
+                    if prefix is None
+                    else codecs.getincrementaldecoder("utf-8")().decode(
+                        prefix, final=False
+                    )[:100]
+                )
+                evidence_by_context[context_id] = []
+                summaries[context_id] = {
+                    key: anchor[key]
+                    for key in (
+                        "context_id",
+                        "positive_snapshot_id",
+                        "negative_snapshot_id",
+                    )
+                }
+                summaries[context_id]["negative_raw"] = {
+                    "state": (
+                        "unknown"
+                        if preview is None
+                        else "explicit_empty" if not preview else "known"
+                    ),
+                    "preview": preview,
+                    "truncated": (
+                        False
+                        if preview is None
+                        else len(preview.encode("utf-8")) < anchor["byte_length"]
+                    ),
+                    "evidence": evidence_by_context[context_id],
+                }
+            negative = json.loads(anchor["evidence_json"])["negative"]
+            evidence = (
+                negative["raw"]
+                if negative is not None
+                else dict(provenance="unknown", source="unknown", reason="missing")
+            )
+            evidence_by_context[context_id].append(
+                {
+                    "run_id": anchor["run_id"],
+                    "anchor_id": anchor["anchor_id"],
+                    **evidence,
+                }
+            )
+        # Context order follows its first chronological matching run, then anchor.
+        ordered = {}
+        for row in rows:
+            for anchor in row["matching_anchors"]:
+                ordered.setdefault(
+                    anchor["context_id"], summaries[anchor["context_id"]]
+                )
+        result["items"] = rows
+        result["context_summaries"] = list(ordered.values())
+        return result
+
+    def get_newest_pair_run(self, positive_id: int, pair_id: int) -> dict | None:
+        """The newest matching run whose current saved-pair hint names `pair_id`.
+
+        None when the positive is not an eligible saved prompt or the pair is
+        not one of its pairs; ``{"run": None}`` when no recorded run names it.
+        Only a hint captured in this epoch and still resolving to the same pair
+        counts, as in ``get_history_run``.
+        """
+        conn = self._history_connection()
+        saved = conn.execute(
+            "SELECT CASE WHEN typeof(p.text)='text' AND "
+            "length(CAST(p.text AS BLOB))<=65536 THEN p.text ELSE NULL END AS text "
+            f"FROM prompts p WHERE p.id=? AND ({_ELIGIBLE})",
+            (positive_id,),
+        ).fetchone()
+        if saved is None or (
+            conn.execute(
+                "SELECT 1 FROM prompt_pairs WHERE id=? AND positive_prompt_id=?",
+                (pair_id, positive_id),
+            ).fetchone()
+            is None
+        ):
+            return None
+        if saved["text"] is None:
+            return {"run": None}
+        epoch = self.get_execution_epoch()
+        candidates = conn.execute(
+            "SELECT r.run_id,rc.anchor_id,rc.context_id,"
+            "c.positive_snapshot_id,c.negative_snapshot_id "
+            "FROM execution_runs r JOIN execution_run_contexts rc "
+            "ON rc.run_id=r.run_id JOIN execution_contexts c "
+            "ON c.context_id=rc.context_id JOIN execution_prompt_snapshots s "
+            "ON s.snapshot_id=c.positive_snapshot_id "
+            "WHERE s.raw_text_id=? AND rc.pair_id=? AND rc.reference_epoch=? "
+            f"ORDER BY {_ORDER},rc.anchor_id COLLATE BINARY LIMIT ?",
+            (
+                content_digest("text", (saved["text"],)),
+                pair_id,
+                epoch,
+                _PAIR_RUN_CANDIDATES,
+            ),
+        ).fetchall()
+        for row in candidates:
+            positive = _values(conn, row["positive_snapshot_id"])
+            negative = _values(conn, row["negative_snapshot_id"])
+            if positive is None or positive["raw"] != saved["text"]:
+                continue
+            hints = _hints(conn, positive["raw"], negative["raw"] if negative else None)
+            if hints[2] == pair_id:
+                return {
+                    "run": {
+                        key: row[key] for key in ("run_id", "anchor_id", "context_id")
+                    }
+                }
+        return {"run": None}
+
+    def get_history_run(self, run_id: str) -> dict | None:
+        conn = self._history_connection()
+        run = self.get_execution_run(run_id)
+        if run is None:
+            return None
+        capture = self.get_execution_contexts(run_id)
+        snapshots = {
+            row["anchor_id"]: row
+            for row in conn.execute(
+                "SELECT rc.anchor_id,c.positive_snapshot_id,c.negative_snapshot_id,"
+                "p.raw_text_id AS positive_raw_text_id "
+                "FROM execution_run_contexts rc JOIN execution_contexts c "
+                "ON c.context_id=rc.context_id LEFT JOIN execution_prompt_snapshots p "
+                "ON p.snapshot_id=c.positive_snapshot_id WHERE rc.run_id=?",
+                (run_id,),
+            )
+        }
+        contexts = capture.pop("contexts")
+        for context in contexts:
+            row = snapshots[context["anchor_id"]]
+            context.update(
+                {
+                    key: row[key]
+                    for key in (
+                        "positive_snapshot_id",
+                        "negative_snapshot_id",
+                        "positive_raw_text_id",
+                    )
+                }
+            )
+            pair = context["pair_id"]
+            context["navigation"] = {
+                "positive_available": context["positive_prompt_id"] is not None
+                and conn.execute(
+                    f"SELECT 1 FROM prompts p WHERE p.id=? AND ({_ELIGIBLE})",
+                    (context["positive_prompt_id"],),
+                ).fetchone()
+                is not None,
+                "negative_available": context["negative_prompt_id"] is not None,
+                "pair_available": pair is not None
+                and conn.execute(
+                    "SELECT 1 FROM prompt_pairs WHERE id=? AND archived=0", (pair,)
+                ).fetchone()
+                is not None,
+            }
+        return dict(
+            run=run,
+            context_capture=capture,
+            contexts=contexts,
+            output_summary=_output_summary(conn, run_id),
+        )
+
+    def list_history_outputs(
+        self,
+        run_id: str,
+        *,
+        limit: int = 40,
+        after: OutputPosition | None = None,
+    ) -> dict | None:
+        conn = self._history_connection()
+        _limit(limit)
+        _validate_output_position(after)
+        # Validate header/orphans even when the requested parent does not exist.
+        capture, count = _output_header(conn, run_id)
+        if self.get_execution_run(run_id) is None:
+            return None
+        fingerprint = _fingerprint(capture, count)
+        if after is not None and after.capture_fingerprint != fingerprint:
+            raise HistoryCaptureChanged("History output capture changed")
+        where = "run_id=?"
+        params = [run_id]
+        if after is not None:
+            where += " AND (output_node_id COLLATE BINARY,image_ordinal)>(?,?)"
+            params.extend(after.after)
+        rows = list(
+            conn.execute(
+                f"SELECT {_REFERENCE_COLUMNS} FROM execution_run_outputs WHERE {where} "
+                "ORDER BY output_node_id COLLATE BINARY,image_ordinal LIMIT ?",
+                (*params, limit + 1),
+            )
+        )
+        more = len(rows) > limit
+        rows = rows[:limit]
+        position = (
+            OutputPosition(
+                fingerprint, (rows[-1]["output_node_id"], rows[-1]["image_ordinal"])
+            )
+            if more
+            else None
+        )
+        return dict(
+            capture=capture,
+            retained_occurrence_count=count,
+            list_status=_list_status(capture),
+            capture_fingerprint=fingerprint,
+            items=[_receipt(run_id, row) for row in rows],
+            has_more=more,
+            next_position=position,
+        )
+
+    def get_history_occurrence(
+        self, run_id: str, node: str, ordinal: int
+    ) -> dict | None:
+        conn = self._history_connection()
+        capture, _ = _output_header(conn, run_id)
+        row = conn.execute(
+            f"SELECT {_REFERENCE_COLUMNS} FROM execution_run_outputs "
+            "WHERE run_id=? AND output_node_id=? COLLATE BINARY AND image_ordinal=?",
+            (run_id, node, ordinal),
+        ).fetchone()
+        return (
+            dict(capture=capture, receipt=_receipt(run_id, row))
+            if row is not None
+            else None
+        )

database/image_favorites.py

@@ -0,0 +1,152 @@
+"""Persisted image favorites, keyed by gallery root + relative path.
+
+The root key is the root's real, case-normalized path and the relative key is
+case-normalized the same way (see ``py/api/images.py``); both stay
+server-side. Folder names alone are not unique across roots, so a favorite
+follows the directory itself when the configured roots are reordered.
+Optional in old backups, required once a database is current (see
+``library_schema.py``).
+"""
+
+import re
+import sqlite3
+from typing import Iterable
+
+TABLE = "image_favorites"
+MAX_KEY_LENGTH = 4096
+_LOOKUP_CHUNK = 500
+_DDL = (
+    f"CREATE TABLE {TABLE} (root_key TEXT NOT NULL, relative_path TEXT NOT NULL, "
+    "created_at TEXT NOT NULL, PRIMARY KEY(root_key, relative_path))"
+)
+_AUTOINDEX = f"sqlite_autoindex_{TABLE}_1"
+_COLUMNS = (
+    ("root_key", "TEXT", 1, 1),
+    ("relative_path", "TEXT", 1, 2),
+    ("created_at", "TEXT", 1, 0),
+)
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql).lower()
+
+
+def image_favorites_schema_error(
+    conn: sqlite3.Connection, *, allow_missing: bool = True
+) -> str | None:
+    """Reject unsupported favorites schemas and invalid persisted values.
+
+    Reads only. A missing table is acceptable for backups that predate it
+    unless ``allow_missing`` is false.
+    """
+    objects = conn.execute(
+        "SELECT type,name,sql FROM sqlite_master WHERE name=? "
+        "OR (type IN ('index','trigger') AND tbl_name=?)",
+        (TABLE, TABLE),
+    ).fetchall()
+    if not objects:
+        return None if allow_missing else "Missing image favorites schema"
+    tables = [sql for kind, _name, sql in objects if kind == "table"]
+    others = {(kind, name, sql) for kind, name, sql in objects if kind != "table"}
+    if len(tables) != 1 or others != {("index", _AUTOINDEX, None)}:
+        return "Image favorites name is reserved by an unsupported schema object"
+    if _compact(tables[0] or "") != _compact(_DDL):
+        return "Image favorites has an invalid table definition"
+    columns = tuple(
+        (row[1], row[2].upper(), row[3], row[5])
+        for row in conn.execute(f"PRAGMA table_info({TABLE})")
+    )
+    if columns != _COLUMNS:
+        return "Image favorites has invalid columns"
+    # Bytes bound characters from above (UTF-8 is at most 4 bytes each).
+    limit = MAX_KEY_LENGTH * 4
+    if conn.execute(
+        f"SELECT 1 FROM {TABLE} WHERE typeof(root_key) != 'text' "
+        "OR typeof(relative_path) != 'text' OR typeof(created_at) != 'text' "
+        f"OR length(CAST(root_key AS BLOB)) > {limit} "
+        f"OR length(CAST(relative_path AS BLOB)) > {limit} "
+        "OR length(CAST(created_at AS BLOB)) > 64 LIMIT 1"
+    ).fetchone():
+        return "Image favorites contains invalid values"
+    for root, path in conn.execute(
+        f"SELECT CAST(root_key AS BLOB), CAST(relative_path AS BLOB) FROM {TABLE} "
+        f"WHERE length(CAST(root_key AS BLOB)) > {MAX_KEY_LENGTH} "
+        f"OR length(CAST(relative_path AS BLOB)) > {MAX_KEY_LENGTH}"
+    ):
+        for raw in (root, path):
+            if len(bytes(raw).decode("utf-8", "replace")) > MAX_KEY_LENGTH:
+                return "Image favorites contains invalid values"
+    return None
+
+
+def migrate_image_favorites(conn: sqlite3.Connection) -> None:
+    """Install the table atomically; validate before and after creating it."""
+    conn.execute("SAVEPOINT image_favorites_schema")
+    try:
+        error = image_favorites_schema_error(conn)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute(_DDL.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        error = image_favorites_schema_error(conn, allow_missing=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute("RELEASE image_favorites_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO image_favorites_schema")
+        conn.execute("RELEASE image_favorites_schema")
+        raise
+
+
+class ImageFavoriteOperations:
+    """PromptDatabase mixin: set, clear and look up image favorites."""
+
+    def set_image_favorite(self, root_key: str, relative_path: str) -> bool:
+        """Mark one file as a favorite; repeating it is a no-op. Returns True."""
+        from .models import utc_now_iso  # models imports this module's schema
+
+        with self.model.get_connection() as conn:
+            conn.execute(
+                "INSERT OR IGNORE INTO image_favorites VALUES (?, ?, ?)",
+                (root_key, relative_path, utc_now_iso()),
+            )
+        return True
+
+    def clear_image_favorite(self, root_key: str, relative_path: str) -> bool:
+        """Remove one favorite; True when a row was removed."""
+        with self.model.get_connection() as conn:
+            cursor = conn.execute(
+                "DELETE FROM image_favorites WHERE root_key=? AND relative_path=?",
+                (root_key, relative_path),
+            )
+        return cursor.rowcount > 0
+
+    def get_image_favorites(
+        self, root_key: str, relative_paths: Iterable[str]
+    ) -> set[str]:
+        """The subset of *relative_paths* favorited under *root_key*."""
+        wanted = list(dict.fromkeys(relative_paths))
+        found: set[str] = set()
+        conn = self.model.get_connection()
+        for start in range(0, len(wanted), _LOOKUP_CHUNK):
+            chunk = wanted[start : start + _LOOKUP_CHUNK]
+            marks = ",".join("?" * len(chunk))
+            rows = conn.execute(  # only "?" placeholders are interpolated
+                "SELECT relative_path FROM image_favorites WHERE root_key=? "
+                f"AND relative_path IN ({marks})",  # nosec B608
+                (root_key, *chunk),
+            )
+            found.update(path for (path,) in rows)
+        return found
+
+    def get_root_favorites(self, root_key: str) -> set[str]:
+        """Every relative key favorited under *root_key* (a primary-key range).
+
+        The Favorites listing (``favorites=1``) intersects these with the
+        root's scan, so its size follows the root's favorites, not its files.
+        """
+        conn = self.model.get_connection()
+        rows = conn.execute(
+            "SELECT relative_path FROM image_favorites WHERE root_key=?",
+            (root_key,),
+        )
+        return {path for (path,) in rows}

database/library.py

@@ -0,0 +1,684 @@
+"""Read-only positive-parent views over existing prompt and pair tables."""
+
+import datetime
+import sqlite3
+from collections.abc import Iterator
+from contextlib import contextmanager
+
+from .library_tokens import library_edit_token
+
+_ELIGIBLE = """(EXISTS(SELECT 1 FROM prompt_pairs ep WHERE ep.positive_prompt_id=p.id)
+ OR EXISTS(SELECT 1 FROM generated_images eg WHERE eg.prompt_id=p.id)
+ OR NOT EXISTS(SELECT 1 FROM prompt_pairs en WHERE en.negative_prompt_id=p.id)
+ OR EXISTS(SELECT 1 FROM prompt_library_state es WHERE es.prompt_id=p.id))"""
+_FROM = "prompts p LEFT JOIN prompt_library_state ls ON ls.prompt_id=p.id"
+# The characters JavaScript's String.prototype.trim() removes, so a title the
+# display treats as blank also falls back to the prompt text for name sorting.
+_TRIM_CODES = (9, 10, 11, 12, 13, 32, 160, 5760, *range(8192, 8203))
+_TRIM_CODES += (8232, 8233, 8239, 8287, 12288, 65279)
+_BLANK = f"char({','.join(map(str, _TRIM_CODES))})"
+_JS_TRIM = "".join(map(chr, _TRIM_CODES))
+# The Library's display title: a stored title, else the prompt's first line.
+DISPLAY_TITLE_LIMIT = 80
+
+
+def _utf16_units(text: str) -> int:
+    """JavaScript ``length``: characters beyond the BMP count as two units."""
+    return len(text) + sum(1 for ch in text if ord(ch) > 0xFFFF)
+
+
+def _utf16_prefix(text: str, units: int) -> str:
+    """The longest prefix within ``units`` UTF-16 units, never splitting a pair."""
+    used = 0
+    for index, ch in enumerate(text):
+        used += 2 if ord(ch) > 0xFFFF else 1
+        if used > units:
+            return text[:index]
+    return text
+
+
+def titled(stored_titles: tuple, text: object) -> tuple[str | None, str | None]:
+    """Return ``(title, source)``: a stored title, else the text's first line.
+
+    ``source`` is ``"stored"`` for the first non-blank stored title (blank as
+    in JavaScript's trim, like the Library view), ``"text"`` for the first line,
+    and None when both are empty. The first line is trimmed and, past
+    DISPLAY_TITLE_LIMIT UTF-16 units, cut with an ellipsis like the Library's
+    ``libraryTitle()`` (never inside a surrogate pair).
+    """
+    for title in stored_titles:
+        if isinstance(title, str) and title.strip(_JS_TRIM):
+            return title, "stored"
+    if not isinstance(text, str):
+        return None, None
+    first = text.split("\n")[0].strip(_JS_TRIM)
+    if _utf16_units(first) > DISPLAY_TITLE_LIMIT:
+        first = _utf16_prefix(first, DISPLAY_TITLE_LIMIT) + "\u2026"
+    return (first, "text") if first else (None, None)
+
+
+def display_title(stored_titles: tuple, text: object) -> str | None:
+    """The Library display title of ``titled``, without its source."""
+    return titled(stored_titles, text)[0]
+
+
+def utc_iso(value: object) -> str | None:
+    """A stored timestamp as UTC ISO text, or None when absent or unreadable.
+
+    Usage writes are already UTC ISO; a naive value (SQLite's CURRENT_TIMESTAMP
+    form) is read as UTC, and an offset is converted to UTC. One bad row must
+    not fail a whole Library read, so a value past the datetime range after
+    conversion (OverflowError) is unreadable too. Python 3.11+ fromisoformat
+    accepts more ISO forms than 3.10 (e.g. a 2-digit fraction); every form the
+    app writes (utc_now_iso, the 3.2.4 backfill, CURRENT_TIMESTAMP, Z) parses
+    the same on both, so only hand-edited values can differ by version.
+    """
+    if not isinstance(value, str):
+        return None
+    text = value.strip()
+    if text.endswith(("Z", "z")):
+        text = text[:-1] + "+00:00"
+    try:
+        stamp = datetime.datetime.fromisoformat(text)
+        if stamp.tzinfo is None:
+            stamp = stamp.replace(tzinfo=datetime.timezone.utc)
+        return stamp.astimezone(datetime.timezone.utc).isoformat()
+    except (ValueError, OverflowError):
+        return None
+
+
+_NAME = (
+    f"LOWER(CASE WHEN TRIM(COALESCE(ls.title,''),{_BLANK})!='' "
+    "THEN ls.title ELSE p.text END)"
+)
+_ORDERS = {
+    "newest": ("COALESCE(julianday(p.created_at),0)", "DESC"),
+    "oldest": ("COALESCE(julianday(p.created_at),0)", "ASC"),
+    "name": (_NAME, "ASC"),
+    "rating": ("COALESCE(p.rating,0)", "DESC"),
+}
+# Pair archive scopes for variant and image reads, as conditions on alias pp.
+ARCHIVE_SCOPES = {
+    "active": " AND pp.archived=0",
+    "archived": " AND pp.archived=1",
+    "all": "",
+}
+
+
+def _escape(value: str) -> str:
+    return value.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+
+
+def _limit(value: int) -> None:
+    if type(value) is not int or not 1 <= value <= 200:
+        raise ValueError("limit must be between 1 and 200")
+
+
+def _archive(value: str) -> str:
+    """Return the pair-archive condition for one scope (alias ``pp``)."""
+    if not isinstance(value, str) or value not in ARCHIVE_SCOPES:
+        raise ValueError("archive must be one of active, archived or all")
+    return ARCHIVE_SCOPES[value]
+
+
+def _filter(text: str, category: str | None, tags: list[str]) -> tuple[str, list]:
+    if not isinstance(text, str) or len(text) > 500:
+        raise ValueError("Search text must be a string of at most 500 characters")
+    if category is not None and (not isinstance(category, str) or len(category) > 200):
+        raise ValueError("Invalid library category")
+    if (
+        not isinstance(tags, list)
+        or len(tags) > 30
+        or any(not isinstance(tag, str) or not tag or len(tag) > 200 for tag in tags)
+    ):
+        raise ValueError("Invalid library tags")
+    parent, pair, parent_params, pair_params = [], ["pp.archived=0"], [], []
+    if text:
+        parent.append("(p.text LIKE ? ESCAPE '\\' OR ls.title LIKE ? ESCAPE '\\')")
+        parent_params.extend([f"%{_escape(text)}%"] * 2)
+        pair.append(
+            "(p.text LIKE ? ESCAPE '\\' OR ls.title LIKE ? ESCAPE '\\' "
+            "OR n.text LIKE ? ESCAPE '\\' OR pp.title LIKE ? ESCAPE '\\')"
+        )
+        pair_params.extend([f"%{_escape(text)}%"] * 4)
+    if category is not None:
+        parent.append("p.category=?")
+        pair.append("pp.category=?")
+        parent_params.append(category)
+        pair_params.append(category)
+    for tag in tags:
+        parent.append(
+            "EXISTS(SELECT 1 FROM prompt_tags pt JOIN tags t ON t.id=pt.tag_id "
+            "WHERE pt.prompt_id=p.id AND t.name=?)"
+        )
+        pair.append(
+            "EXISTS(SELECT 1 FROM prompt_pair_tags pt JOIN tags t ON t.id=pt.tag_id "
+            "WHERE pt.pair_id=pp.id AND t.name=?)"
+        )
+        parent_params.append(tag)
+        pair_params.append(tag)
+    condition = (
+        "("
+        + (" AND ".join(parent) or "1=1")
+        + (
+            ") OR EXISTS(SELECT 1 FROM prompt_pairs pp "
+            "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id "
+            "WHERE pp.positive_prompt_id=p.id AND "
+        )
+        + " AND ".join(pair)
+        + ")"
+    )
+    return f"({_ELIGIBLE}) AND ({condition})", [*parent_params, *pair_params]
+
+
+class LibraryOperations:
+    """Group reusable negative variants without changing legacy prompt queries.
+
+    Each public read runs inside ``library_snapshot``; the ``_``-prefixed
+    helpers take that snapshot's connection and epoch so routes can share it.
+    """
+
+    @contextmanager
+    def library_snapshot(
+        self, *, include_ceiling: bool = False
+    ) -> Iterator[tuple[sqlite3.Connection, str, int | None]]:
+        """Hold one read transaction, reading the epoch (and ceiling) first."""
+        conn = self.model.get_connection()
+        if conn.in_transaction:
+            raise sqlite3.ProgrammingError("Library reads require their own snapshot")
+        conn.execute("BEGIN")
+        try:
+            row = conn.execute(
+                "SELECT storage_epoch,"
+                + (
+                    "(SELECT COALESCE(MAX(id),0) FROM prompts)"
+                    if include_ceiling
+                    else "NULL"
+                )
+                + " FROM execution_history_metadata WHERE singleton_id=1"
+            ).fetchone()
+            if row is None:
+                raise sqlite3.DatabaseError("Missing execution history metadata")
+            yield conn, row[0], row[1]
+        finally:
+            if conn.in_transaction:
+                conn.rollback()
+
+    def _library_parents(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        where: str,
+        params: list,
+        suffix: str = "",
+        sort_expression: str | None = None,
+    ) -> list[dict]:
+        rows = conn.execute(
+            "SELECT p.id,p.text,ls.title,p.category,p.rating,p.notes,p.created_at,"
+            "p.run_count,p.last_used_at,"
+            + (f"{sort_expression} AS _sort," if sort_expression else "")
+            + "(SELECT COUNT(*) FROM prompt_pairs v WHERE v.positive_prompt_id=p.id "
+            "AND v.archived=0) AS variant_count,"
+            "(SELECT COUNT(*) FROM generated_images g "
+            "WHERE g.prompt_id=p.id) AS image_count "
+            "FROM " + _FROM + " WHERE " + where + suffix,
+            params,
+        ).fetchall()
+        result = [{**dict(row), "tags": []} for row in rows]
+        if result:
+            by_id = {row["id"]: row for row in result}
+            marks = ",".join("?" for _ in result)
+            for row in conn.execute(
+                "SELECT pt.prompt_id,t.name FROM prompt_tags pt "
+                "JOIN tags t ON t.id=pt.tag_id "
+                f"WHERE pt.prompt_id IN ({marks}) ORDER BY t.name",
+                list(by_id),
+            ):
+                by_id[row["prompt_id"]]["tags"].append(row["name"])
+        for row in result:
+            # Usage is shown, never compared: the token hashes owner fields only.
+            row["run_count"] = row["run_count"] if type(row["run_count"]) is int else 0
+            row["last_used_at"] = utc_iso(row["last_used_at"])
+            row["edit_token"] = library_edit_token(epoch, row)
+        return result
+
+    def _list_library(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        ceiling: int,
+        *,
+        text: str = "",
+        category: str | None = None,
+        tags: list[str] | None = None,
+        sort: str = "newest",
+        limit: int = 40,
+        after: dict | None = None,
+    ) -> dict:
+        _limit(limit)
+        if sort not in _ORDERS:
+            raise ValueError("Unknown library sort order")
+        where, params = _filter(text, category, [] if tags is None else tags)
+        if after is not None:
+            if (
+                not isinstance(after, dict)
+                or set(after) != {"ceiling", "value", "id"}
+                or type(after["ceiling"]) is not int
+                or type(after["id"]) is not int
+            ):
+                raise ValueError("Invalid library cursor")
+            ceiling = after["ceiling"]
+        where += " AND p.id<=?"
+        params.append(ceiling)
+        # One SELECT counts the filtered parents and their active variants.
+        total, variant_total = conn.execute(
+            "SELECT COUNT(*),COALESCE(SUM((SELECT COUNT(*) FROM prompt_pairs v "
+            "WHERE v.positive_prompt_id=p.id AND v.archived=0)),0) FROM "
+            + _FROM
+            + " WHERE "
+            + where,
+            params,
+        ).fetchone()
+        expression, direction = _ORDERS[sort]
+        if after is not None:
+            operator = ">" if direction == "ASC" else "<"
+            where += (
+                f" AND ({expression}{operator}? OR "
+                f"({expression}=? AND p.id{operator}?))"
+            )
+            params.extend([after["value"], after["value"], after["id"]])
+        rows = self._library_parents(
+            conn,
+            epoch,
+            where,
+            [*params, limit + 1],
+            f" ORDER BY {expression} {direction},p.id {direction} LIMIT ?",
+            sort_expression=expression,
+        )
+        has_more = len(rows) > limit
+        rows = rows[:limit]
+        next_key = None
+        if has_more:
+            next_key = {
+                "ceiling": ceiling,
+                "value": rows[-1]["_sort"],
+                "id": rows[-1]["id"],
+            }
+        for row in rows:
+            row.pop("_sort")
+        return {
+            "library_epoch": epoch,
+            "items": rows,
+            "total": total,
+            "variant_total": variant_total,
+            "has_more": has_more,
+            "next_key": next_key,
+        }
+
+    def list_library(
+        self,
+        *,
+        text: str = "",
+        category: str | None = None,
+        tags: list[str] | None = None,
+        sort: str = "newest",
+        limit: int = 40,
+        after: dict | None = None,
+    ) -> dict:
+        with self.library_snapshot(include_ceiling=True) as (conn, epoch, ceiling):
+            return self._list_library(
+                conn,
+                epoch,
+                ceiling,
+                text=text,
+                category=category,
+                tags=tags,
+                sort=sort,
+                limit=limit,
+                after=after,
+            )
+
+    def _is_library_positive(self, conn: sqlite3.Connection, positive_id: int) -> bool:
+        return (
+            conn.execute(
+                "SELECT 1 FROM prompts p WHERE p.id=? AND " + _ELIGIBLE, (positive_id,)
+            ).fetchone()
+            is not None
+        )
+
+    def _get_library_positive(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        positive_id: int,
+        *,
+        variant_limit: int = 40,
+        variant_after: int | None = None,
+        archive: str = "active",
+    ) -> dict | None:
+        _limit(variant_limit)
+        scope = _archive(archive)
+        if variant_after is not None and (
+            type(variant_after) is not int or variant_after <= 0
+        ):
+            raise ValueError("Invalid variant cursor")
+        rows = self._library_parents(
+            conn, epoch, _ELIGIBLE + " AND p.id=?", [positive_id]
+        )
+        if not rows:
+            return None
+        positive = rows[0]
+        positive["archived_variant_count"] = conn.execute(
+            "SELECT COUNT(*) FROM prompt_pairs "
+            "WHERE positive_prompt_id=? AND archived=1",
+            (positive_id,),
+        ).fetchone()[0]
+        variant_total = {
+            "active": positive["variant_count"],
+            "archived": positive["archived_variant_count"],
+            "all": positive["variant_count"] + positive["archived_variant_count"],
+        }[archive]
+        where = "pp.positive_prompt_id=?" + scope
+        params = [positive_id]
+        if variant_after is not None:
+            where += " AND pp.id<?"
+            params.append(variant_after)
+        variants = [
+            dict(row)
+            for row in conn.execute(
+                "SELECT pp.id,COALESCE(n.text,'') AS negative_text,"
+                "pp.category,pp.title,pp.archived,"
+                "pp.negative_prompt_id IS NULL AS explicit_empty,pp.run_count,"
+                "(SELECT COUNT(DISTINCT image_id) FROM image_prompt_pairs ip "
+                "WHERE ip.pair_id=pp.id) AS image_count "
+                "FROM prompt_pairs pp "
+                "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id "
+                "WHERE " + where + " ORDER BY pp.id DESC LIMIT ?",
+                [*params, variant_limit + 1],
+            )
+        ]
+        variants_has_more = len(variants) > variant_limit
+        variants = variants[:variant_limit]
+        by_id = {row["id"]: row for row in variants}
+        for row in variants:
+            row["tags"] = []
+            row["archived"] = bool(row["archived"])
+            row["explicit_empty"] = bool(row["explicit_empty"])
+            # The history lane's "N runs" (#249): queue-hook pair uses, 0 when unset.
+            row["run_count"] = row["run_count"] if type(row["run_count"]) is int else 0
+        if by_id:
+            marks = ",".join("?" for _ in by_id)
+            for row in conn.execute(
+                "SELECT pt.pair_id,t.name FROM prompt_pair_tags pt "
+                "JOIN tags t ON t.id=pt.tag_id "
+                f"WHERE pt.pair_id IN ({marks}) ORDER BY t.name",
+                list(by_id),
+            ):
+                by_id[row["pair_id"]]["tags"].append(row["name"])
+        unassigned = conn.execute(
+            "SELECT COUNT(*) FROM generated_images g WHERE g.prompt_id=? "
+            "AND NOT EXISTS(SELECT 1 FROM image_prompt_pairs ip "
+            "WHERE ip.image_id=g.id)",
+            (positive_id,),
+        ).fetchone()[0]
+        return {
+            "library_epoch": epoch,
+            "positive": positive,
+            "variants": variants,
+            "unassigned_image_count": unassigned,
+            "variant_total": variant_total,
+            "variants_has_more": variants_has_more,
+            "variants_next_key": variants[-1]["id"] if variants_has_more else None,
+        }
+
+    def get_library_positive(
+        self,
+        positive_id: int,
+        *,
+        variant_limit: int = 40,
+        variant_after: int | None = None,
+        archive: str = "active",
+    ) -> dict | None:
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            return self._get_library_positive(
+                conn,
+                epoch,
+                positive_id,
+                variant_limit=variant_limit,
+                variant_after=variant_after,
+                archive=archive,
+            )
+
+    def _get_library_options(self, conn: sqlite3.Connection, epoch: str) -> dict:
+        rows = conn.execute(
+            "SELECT category FROM (SELECT p.category AS category "
+            "FROM prompts p WHERE "
+            + _ELIGIBLE
+            + (" UNION SELECT category FROM prompt_pairs " "WHERE archived=0) ")
+            + "WHERE category IS NOT NULL AND TRIM(category)!='' "
+            "ORDER BY category COLLATE NOCASE,category"
+        ).fetchall()
+        return {
+            "library_epoch": epoch,
+            "categories": [row["category"] for row in rows],
+        }
+
+    def get_library_options(self) -> dict:
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            return self._get_library_options(conn, epoch)
+
+    def get_library_previews(self, positive_ids: list[int]) -> dict[int, list[dict]]:
+        """Read at most two links per parent in one bounded page query."""
+        if (
+            not isinstance(positive_ids, list)
+            or len(positive_ids) > 200
+            or any(type(value) is not int or value <= 0 for value in positive_ids)
+        ):
+            raise ValueError("Invalid preview parent IDs")
+        if not positive_ids:
+            return {}
+        result = {value: [] for value in positive_ids}
+        marks = ",".join("?" for _ in positive_ids)
+        rows = self.model.get_connection().execute(
+            "WITH ranked AS (SELECT id,prompt_id,image_path,ROW_NUMBER() OVER "
+            "(PARTITION BY prompt_id ORDER BY id DESC) AS position "
+            f"FROM generated_images WHERE prompt_id IN ({marks})) "
+            "SELECT id,prompt_id,image_path FROM ranked WHERE position<=2 "
+            "ORDER BY prompt_id,id DESC",
+            positive_ids,
+        )
+        for row in rows:
+            result[row["prompt_id"]].append(dict(row))
+        return result
+
+    def _list_library_images(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        positive_id: int,
+        *,
+        pair_id: int | None = None,
+        limit: int = 40,
+        after: int | None = None,
+        archive: str = "active",
+    ) -> dict:
+        _limit(limit)
+        scope = _archive(archive)
+        if after is not None and (type(after) is not int or after <= 0):
+            raise ValueError("Invalid image cursor")
+        if (
+            pair_id is not None
+            and not conn.execute(
+                "SELECT 1 FROM prompt_pairs pp WHERE pp.id=? "
+                "AND pp.positive_prompt_id=?" + scope,
+                (pair_id, positive_id),
+            ).fetchone()
+        ):
+            if conn.execute(
+                "SELECT 1 FROM prompt_pairs WHERE id=? AND positive_prompt_id=?",
+                (pair_id, positive_id),
+            ).fetchone():
+                raise ValueError("Variant is not in the requested archive scope")
+            raise ValueError("Variant does not belong to this positive prompt")
+        where, params = "g.prompt_id=?", [positive_id]
+        if pair_id is not None:
+            where += (
+                " AND EXISTS(SELECT 1 FROM image_prompt_pairs ip "
+                "WHERE ip.image_id=g.id AND ip.pair_id=?)"
+            )
+            params.append(pair_id)
+        total = conn.execute(
+            "SELECT COUNT(*) FROM generated_images g WHERE " + where, params
+        ).fetchone()[0]
+        if after is not None:
+            where += " AND g.id<?"
+            params.append(after)
+        rows = conn.execute(
+            "SELECT g.id,g.image_path FROM generated_images g WHERE "
+            + where
+            + " ORDER BY g.id DESC LIMIT ?",
+            [*params, limit + 1],
+        ).fetchall()
+        return {
+            "library_epoch": epoch,
+            "images": [dict(row) for row in rows[:limit]],
+            "total": total,
+            "has_more": len(rows) > limit,
+            "next_key": rows[limit - 1]["id"] if len(rows) > limit else None,
+        }
+
+    def list_library_images(
+        self,
+        positive_id: int,
+        *,
+        pair_id: int | None = None,
+        limit: int = 40,
+        after: int | None = None,
+        archive: str = "active",
+    ) -> dict:
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            return self._list_library_images(
+                conn,
+                epoch,
+                positive_id,
+                pair_id=pair_id,
+                limit=limit,
+                after=after,
+                archive=archive,
+            )
+
+    def export_library(self) -> dict:
+        """Read every portable owner and pair from one snapshot, writing nothing.
+
+        Local IDs order the rows internally and never appear in the document.
+        """
+        with self.library_snapshot() as (conn, _epoch, _ceiling):
+            owners = {}
+            order = []
+            for row in conn.execute(
+                "SELECT p.id,p.text,ls.title,p.category,p.rating,p.notes FROM "
+                + _FROM
+                + " WHERE "
+                + _ELIGIBLE
+                + " ORDER BY p.id"
+            ):
+                owners[row["id"]] = {
+                    "positive_text": row["text"],
+                    "title": row["title"],
+                    "category": row["category"],
+                    "tags": [],
+                    "rating": row["rating"],
+                    "notes": row["notes"],
+                    "variants": [],
+                }
+                order.append(row["id"])
+            for row in conn.execute(
+                "SELECT pt.prompt_id,t.name FROM prompt_tags pt "
+                "JOIN tags t ON t.id=pt.tag_id ORDER BY pt.prompt_id,t.name"
+            ):
+                if row["prompt_id"] in owners:
+                    owners[row["prompt_id"]]["tags"].append(row["name"])
+            variants = {}
+            for row in conn.execute(
+                "SELECT pp.id,pp.positive_prompt_id,"
+                "COALESCE(n.text,'') AS negative_text,"
+                "pp.title,pp.category,pp.rating,pp.notes,pp.archived "
+                "FROM prompt_pairs pp "
+                "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id "
+                "ORDER BY pp.id"
+            ):
+                variant = {
+                    "negative_text": row["negative_text"],
+                    "title": row["title"],
+                    "category": row["category"],
+                    "tags": [],
+                    "rating": row["rating"],
+                    "notes": row["notes"],
+                    "archived": bool(row["archived"]),
+                }
+                variants[row["id"]] = variant
+                owners[row["positive_prompt_id"]]["variants"].append(variant)
+            for row in conn.execute(
+                "SELECT pt.pair_id,t.name FROM prompt_pair_tags pt "
+                "JOIN tags t ON t.id=pt.tag_id ORDER BY pt.pair_id,t.name"
+            ):
+                variants[row["pair_id"]]["tags"].append(row["name"])
+            return {
+                "schema_version": 1,
+                "kind": "prompt-manager-library",
+                "positives": [owners[key] for key in order],
+            }
+
+    def library_counts(self) -> dict:
+        """System › Overview counts from one snapshot, writing nothing.
+
+        ``positive_prompts`` and ``negative_variants`` are the unfiltered
+        Library's ``total`` and ``variant_total``: eligible parents and active
+        pairs (every pair's positive is eligible). ``images`` counts indexed
+        ``generated_images`` rows, not files on disk. An image has no rating
+        of its own (it shows its linked prompt's), so ``rated_images`` counts
+        the indexed images whose linked prompt is rated. ``review_pending`` is
+        the Needs review queue's pending total.
+        """
+        with self.library_snapshot() as (conn, _epoch, _ceiling):
+            row = conn.execute(
+                "SELECT (SELECT COUNT(*) FROM prompts p WHERE " + _ELIGIBLE + "),"
+                "(SELECT COUNT(*) FROM prompt_pairs WHERE archived=0),"
+                "(SELECT COUNT(*) FROM generated_images),"
+                "(SELECT COUNT(*) FROM generated_images gi JOIN prompts r "
+                "ON r.id=gi.prompt_id WHERE r.rating IS NOT NULL),"
+                "(SELECT COUNT(*) FROM media_review_items WHERE state='pending')"
+            ).fetchone()
+        return {
+            "positive_prompts": row[0],
+            "negative_variants": row[1],
+            "images": row[2],
+            "rated_images": row[3],
+            "review_pending": row[4],
+        }
+
+    def usage_by_category(self) -> list[dict]:
+        """Runs and average rating of the Library's parents, per category.
+
+        Only eligible parents count, so a negative-only prompt's runs (which
+        repeat its pairs' runs) are never added twice. A blank category groups
+        with no category as ``None``. Unrated parents are left out of the
+        average, which is ``None`` when no parent is rated. Most runs first.
+        """
+        with self.library_snapshot() as (conn, _epoch, _ceiling):
+            rows = conn.execute(
+                "SELECT * FROM (SELECT NULLIF(TRIM(p.category),'') AS category,"
+                "COALESCE(SUM(p.run_count),0) AS runs,AVG(p.rating) AS rating "
+                "FROM prompts p WHERE " + _ELIGIBLE + " GROUP BY 1) "
+                "ORDER BY runs DESC,category IS NULL,category"
+            ).fetchall()
+        return [
+            {
+                "category": row["category"],
+                "runs": row["runs"],
+                "average_rating": (
+                    None if row["rating"] is None else round(row["rating"], 2)
+                ),
+            }
+            for row in rows
+        ]

database/library_commands.py

@@ -0,0 +1,310 @@
+"""Transactional native Library commands guarded by epoch and edit token."""
+
+import re
+import sqlite3
+
+from .library import _ELIGIBLE
+from .models import utc_now_iso
+
+try:
+    from ..utils.validators import (
+        validate_category,
+        validate_prompt_text,
+        validate_tags,
+    )
+except ImportError:
+    from utils.validators import validate_category, validate_prompt_text, validate_tags
+
+METADATA_FIELDS = frozenset({"title", "category", "tags", "rating", "notes"})
+MAX_TEXT_FIELD = 10000
+_UUID = re.compile(r"[0-9a-f]{8}(?:-[0-9a-f]{4}){3}-[0-9a-f]{12}")
+_TOKEN = re.compile(r"[0-9a-f]{64}")
+# Parent metadata stored on the prompt row; title lives in prompt_library_state.
+_PROMPT_COLUMNS = ("category", "rating", "notes")
+
+
+class LibraryCommandError(Exception):
+    """A native command lost a race against storage or another writer.
+
+    ``code`` is a wire code (``metadata-conflict``, ``storage-changed`` or
+    ``vocabulary-conflict`` for a tag or category that changed since it was read),
+    ``epoch`` the storage epoch read inside the transaction and ``positive``
+    the current owner record when it may be disclosed.
+    """
+
+    def __init__(self, code: str, epoch: str, positive: dict | None = None):
+        super().__init__(code)
+        self.code = code
+        self.epoch = epoch
+        self.positive = positive
+
+
+def validate_library_metadata(metadata: dict, *, require_title: bool = False) -> None:
+    """Reject anything but a nonempty allowlisted parent metadata object."""
+    if not isinstance(metadata, dict) or not metadata:
+        raise ValueError("metadata must be a nonempty object")
+    unknown = set(metadata) - METADATA_FIELDS
+    if unknown:
+        raise ValueError("Unknown metadata fields: " + ", ".join(sorted(unknown)))
+    if require_title and (
+        not isinstance(metadata.get("title"), str) or not metadata["title"].strip()
+    ):
+        raise ValueError("title is required")
+    for name in ("title", "notes"):
+        value = metadata.get(name)
+        if value is not None and not isinstance(value, str):
+            raise ValueError(f"{name} must be a string or null")
+        if value is not None and len(value) > MAX_TEXT_FIELD:
+            raise ValueError(f"{name} is too long")
+    if metadata.get("category") is not None:
+        validate_category(metadata["category"])
+    if "tags" in metadata:
+        if not isinstance(metadata["tags"], list):
+            raise ValueError("tags must be a list of strings")
+        validate_tags(metadata["tags"])
+    rating = metadata.get("rating")
+    if rating is not None and (type(rating) is not int or not 1 <= rating <= 5):
+        raise ValueError("rating must be an integer between 1 and 5 or null")
+
+
+def normalize_library_metadata(metadata: dict) -> dict:
+    """Return validated metadata with category trimmed and blank stored as NULL."""
+    if "category" not in metadata:
+        return metadata
+    category = metadata["category"]
+    return {**metadata, "category": (category or "").strip() or None}
+
+
+def normalize_library_tags(tags: list[str]) -> list[str]:
+    """Trim and deduplicate validated tags, keeping first-seen order."""
+    return list(dict.fromkeys(tag.strip() for tag in tags))
+
+
+def _validate_epoch(expected_epoch: str) -> None:
+    if not isinstance(expected_epoch, str) or not _UUID.fullmatch(expected_epoch):
+        raise ValueError("expected_epoch must be a canonical UUID")
+
+
+def _validate_guards(positive_id: int, expected_epoch: str, if_match: str) -> None:
+    if type(positive_id) is not int or not 1 <= positive_id <= 2**63 - 1:
+        raise ValueError("Invalid positive prompt ID")
+    _validate_epoch(expected_epoch)
+    if not isinstance(if_match, str) or not _TOKEN.fullmatch(if_match):
+        raise ValueError("if_match must be 64 lowercase hexadecimal characters")
+
+
+def _validate_negative_text(negative_text: str) -> None:
+    """Require an explicit string; blank deliberately means the empty negative."""
+    if not isinstance(negative_text, str):
+        raise ValueError("negative_text must be explicit; use an empty string for none")
+    if negative_text.strip():
+        validate_prompt_text(negative_text)
+
+
+class LibraryCommandOperations:
+    """Connection-level parent commands mixed into PromptDatabase."""
+
+    def _library_epoch(self, conn: sqlite3.Connection, expected_epoch: str) -> str:
+        row = conn.execute(
+            "SELECT storage_epoch FROM execution_history_metadata WHERE singleton_id=1"
+        ).fetchone()
+        if row is None:
+            raise sqlite3.DatabaseError("Missing execution history metadata")
+        if row[0] != expected_epoch:
+            raise LibraryCommandError("storage-changed", row[0])
+        return row[0]
+
+    def _library_parent(
+        self, conn: sqlite3.Connection, epoch: str, positive_id: int
+    ) -> dict | None:
+        rows = self._library_parents(
+            conn, epoch, _ELIGIBLE + " AND p.id=?", [positive_id]
+        )
+        return rows[0] if rows else None
+
+    def _ensure_library_parent(self, conn: sqlite3.Connection, prompt_id: int) -> None:
+        """Register explicit parent intent once; an existing row is never touched."""
+        if conn.execute(
+            "SELECT 1 FROM prompt_library_state WHERE prompt_id=?", (prompt_id,)
+        ).fetchone():
+            return
+        conn.execute(
+            "INSERT INTO prompt_library_state(prompt_id,title) VALUES(?,NULL)",
+            (prompt_id,),
+        )
+
+    def _apply_library_metadata(
+        self, conn: sqlite3.Connection, current: dict, metadata: dict
+    ) -> None:
+        """Write only fields that differ; stamp ``prompts.updated_at`` once."""
+        prompt_id = current["id"]
+        changed = False
+        if "title" in metadata and metadata["title"] != current["title"]:
+            conn.execute(
+                "UPDATE prompt_library_state SET title=? WHERE prompt_id=?",
+                (metadata["title"], prompt_id),
+            )
+            changed = True
+        columns = {
+            name: metadata[name]
+            for name in _PROMPT_COLUMNS
+            if name in metadata and metadata[name] != current[name]
+        }
+        if "tags" in metadata:
+            tags = normalize_library_tags(metadata["tags"])
+            if set(tags) != set(current["tags"]):
+                self._sync_prompt_tags(conn, prompt_id, tags)
+                changed = True
+        if columns or changed:
+            columns["updated_at"] = utc_now_iso()
+            conn.execute(
+                "UPDATE prompts SET "
+                + ",".join(f"{name}=?" for name in columns)
+                + " WHERE id=?",
+                [*columns.values(), prompt_id],
+            )
+
+    def update_library_parent(
+        self,
+        positive_id: int,
+        *,
+        expected_epoch: str,
+        if_match: str,
+        metadata: dict,
+    ) -> dict | None:
+        """Apply explicit parent metadata if the epoch and edit token still match.
+
+        Returns ``{library_epoch, positive}`` read before commit, or None when
+        the positive is missing or not a Library parent. Raises
+        LibraryCommandError for a rotated epoch or a stale token.
+        """
+        _validate_guards(positive_id, expected_epoch, if_match)
+        validate_library_metadata(metadata)
+        metadata = normalize_library_metadata(metadata)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            current = self._library_parent(conn, epoch, positive_id)
+            if current is None:
+                return None
+            if current["edit_token"] != if_match:
+                raise LibraryCommandError("metadata-conflict", epoch, current)
+            self._ensure_library_parent(conn, positive_id)
+            self._apply_library_metadata(conn, current, metadata)
+            positive = self._library_parent(conn, epoch, positive_id)
+            if positive is None:
+                raise sqlite3.DatabaseError("Library parent vanished during update")
+            return {"library_epoch": epoch, "positive": positive}
+
+    def _recipe_negative(
+        self, conn: sqlite3.Connection, negative_text: str
+    ) -> int | None:
+        if not negative_text.strip():
+            return None
+        return self._pair_component_outcome(conn, negative_text)[0]
+
+    def _recipe_receipt(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        positive_id: int,
+        pair_id: int,
+        *,
+        positive_created: bool,
+        pair_created: bool,
+    ) -> dict:
+        """Read both owners inside the write transaction for the receipt."""
+        positive = self._library_parent(conn, epoch, positive_id)
+        pairs = self._read_pair_rows(conn, "pp.id = ?", [pair_id])
+        if positive is None or not pairs:
+            raise sqlite3.DatabaseError("Library recipe vanished during save")
+        return {
+            "library_epoch": epoch,
+            "positive": positive,
+            "pair": pairs[0],
+            "positive_created": positive_created,
+            "pair_created": pair_created,
+            # Parent metadata is applied only to a newly inserted positive.
+            "parent_metadata_applied": positive_created,
+        }
+
+    def create_library_recipe(
+        self,
+        positive_text: str,
+        negative_text: str,
+        *,
+        expected_epoch: str,
+        parent_metadata: dict,
+    ) -> dict:
+        """Create or reuse a positive and its ordered pair in one transaction.
+
+        Submitted parent metadata applies only when the positive component is
+        newly inserted; a reused component keeps its own metadata. New pairs get
+        empty pair metadata. Reused pairs keep theirs, including archive state.
+        Every success registers the positive as a Library parent.
+        """
+        _validate_epoch(expected_epoch)
+        validate_prompt_text(positive_text)
+        _validate_negative_text(negative_text)
+        validate_library_metadata(parent_metadata, require_title=True)
+        parent_metadata = normalize_library_metadata(parent_metadata)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            positive_id, positive_created = self._pair_component_outcome(
+                conn, positive_text
+            )
+            negative_id = self._recipe_negative(conn, negative_text)
+            pair_id, pair_created = self._save_pair_components(
+                conn, positive_id, negative_id
+            )
+            self._ensure_library_parent(conn, positive_id)
+            if positive_created:
+                current = self._library_parent(conn, epoch, positive_id)
+                if current is None:
+                    raise sqlite3.DatabaseError("New Library parent is not eligible")
+                self._apply_library_metadata(conn, current, parent_metadata)
+            return self._recipe_receipt(
+                conn,
+                epoch,
+                positive_id,
+                pair_id,
+                positive_created=positive_created,
+                pair_created=pair_created,
+            )
+
+    def add_library_negative(
+        self,
+        positive_id: int,
+        negative_text: str,
+        *,
+        expected_epoch: str,
+        if_match: str,
+    ) -> dict | None:
+        """Create or reuse a pair under the exact stored positive.
+
+        The stored positive ID and text are used as they are; only the negative
+        is resolved. Returns None when the positive is missing or not a Library
+        parent, and raises LibraryCommandError for a rotated epoch or a stale
+        token, in both cases before any insertion.
+        """
+        _validate_guards(positive_id, expected_epoch, if_match)
+        _validate_negative_text(negative_text)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            current = self._library_parent(conn, epoch, positive_id)
+            if current is None:
+                return None
+            if current["edit_token"] != if_match:
+                raise LibraryCommandError("metadata-conflict", epoch, current)
+            negative_id = self._recipe_negative(conn, negative_text)
+            pair_id, pair_created = self._save_pair_components(
+                conn, positive_id, negative_id
+            )
+            self._ensure_library_parent(conn, positive_id)
+            return self._recipe_receipt(
+                conn,
+                epoch,
+                positive_id,
+                pair_id,
+                positive_created=False,
+                pair_created=pair_created,
+            )

database/library_move.py

@@ -0,0 +1,805 @@
+"""Move library (#249 S6b): copy the node-folder library, verify it, switch.
+
+``plan_move`` checks a ``utils.library_location`` location can move and
+sizes the copy; ``move_library`` does it, in this order:
+
+1. **Disk check** on the target's volume (the copy, the backups and
+   config.json, plus a margin). The source is never renamed across folders,
+   so another drive is simply a copy.
+2. **Copy** with SQLite's backup API from a private connection holding one
+   read transaction: the per-table counts, the workflow documents' digest
+   and ``PRAGMA data_version`` describe exactly the copied snapshot. The copy
+   is a temp file in the target folder, made self-contained
+   (``journal_mode=DELETE``), fsynced and owner-only.
+3. **Verify**: ``verify_database_file`` (integrity check, schema, every
+   workflow document decompressed against its hash, #315), the same row
+   count in every table and the same documents.
+4. **Copy files**: every pooled safety backup and pre-migration backup,
+   checked by size and SHA-256 after the copy (their times are kept, so the
+   pool prunes in the same order).
+5. **Publish** under the old database's write lock (``BEGIN IMMEDIATE``) and
+   the config.json lock: if any other connection wrote since the snapshot
+   (``data_version``), steps 2–4 run again (``attempts``, then
+   ``library-busy``). Else config.json is copied (owner-only, checked), a
+   ``pending`` marker is written, the temp files take their final names
+   (``os.replace`` inside the target folder), the marker turns ``moved``
+   and ``switch`` points the process at the new library while the old file
+   is still locked, so no write can land in it.
+6. **Retire**: the old file's WAL is checkpointed and the old ``prompts.db``
+   and ``config.json`` are renamed ``*.moved-to-<folder>`` (never deleted;
+   the old backups stay where they are). A file still open elsewhere is left
+   in place and reported (``old-in-use``): the marker already makes a restart
+   use the new library.
+
+Any failure before the marker says ``moved`` removes everything the move
+wrote in the target folder and leaves the old library in use, untouched.
+A crash at any point leaves either the old library in use (the marker is
+absent or ``pending``; the next move replaces its own leftovers) or the new
+one (``moved``). ``MoveFailed`` carries a code and a message without paths.
+"""
+
+import contextlib
+import hashlib
+import json
+import os
+import shutil
+import sqlite3
+import time
+import types
+import uuid
+from pathlib import Path
+from dataclasses import dataclass, field
+from typing import Callable, Dict, List, Optional, Tuple
+
+from .models import BUSY_TIMEOUT_SECONDS, PromptModel
+
+try:
+    from ..utils import library_location as ll
+    from ..utils.logging_config import get_logger
+except ImportError:
+    from utils import library_location as ll
+    from utils.logging_config import get_logger
+
+logger = get_logger("prompt_manager.database.library_move")
+
+TEMP_PREFIX = ".pm-move-"
+MOVED_INFIX = ".moved-to-"
+DISK_MARGIN_BYTES = 16 * 2**20
+DEFAULT_ATTEMPTS = 3
+_COPY_CHUNK = 4 * 2**20
+# Backups that live beside the database and move with it.
+BACKUP_INFIXES = (".backup_", ".pre-workflow-storage_")
+
+MESSAGES = {
+    "not-movable": (
+        "This library is not in the node folder, so there is nothing to move."
+    ),
+    "same-folder": "The new folder is the node folder itself.",
+    "target-invalid": "The new folder cannot hold the library.",
+    "target-has-library": (
+        "The new folder already holds a Prompt Manager library. Nothing was moved."
+    ),
+    "target-conflict": (
+        "The new folder already holds files with the same names. Nothing was moved."
+    ),
+    "insufficient-space": (
+        "There is not enough free space for the copy. Nothing was moved."
+    ),
+    "copy-failed": "The copy could not be written or did not match. Nothing was moved.",
+    "verify-failed": "The copy did not pass verification. Nothing was moved.",
+    "library-busy": (
+        "Prompt Manager kept writing to the library while it was copied. "
+        "Nothing was moved; try again when no workflow is running."
+    ),
+    "switch-failed": (
+        "Prompt Manager could not switch to the verified copy, so the copy was "
+        "removed. Nothing was moved."
+    ),
+}
+
+
+class MoveFailed(Exception):
+    """A move that did not happen (or did not finish), with a coded reason."""
+
+    def __init__(self, code: str, message: Optional[str] = None):
+        self.code = code
+        self.message = message or MESSAGES.get(code, "The library was not moved.")
+        super().__init__(self.message)
+
+
+@dataclass(frozen=True)
+class MovePlan:
+    location: "ll.LibraryLocation"
+    db_path: str
+    source_dir: str
+    target_dir: str
+    config_path: Optional[str]
+    backups: Tuple[str, ...]
+    db_bytes: int
+    files_bytes: int
+
+    @property
+    def required_bytes(self) -> int:
+        return self.db_bytes + self.files_bytes + DISK_MARGIN_BYTES
+
+    @property
+    def target_location(self) -> "ll.LibraryLocation":
+        """Where the library is once moved (an explicit config.json stays)."""
+        source = self.location.target_source or ll.SOURCE_DEFAULT
+        config = (
+            self.location.config_path
+            if self.location.config_explicit
+            else os.path.join(self.target_dir, ll.CONFIG_NAME)
+        )
+        return ll.LibraryLocation(
+            source,
+            os.path.join(self.target_dir, ll.DATABASE_NAME),
+            config,
+            self.target_dir,
+            self.target_dir,
+            None,
+            source,
+            self.location.config_explicit,
+        )
+
+
+@dataclass
+class MoveResult:
+    location: "ll.LibraryLocation"
+    counts: Dict[str, int]
+    files: List[str]
+    renamed: List[str] = field(default_factory=list)
+    retire_problem: Optional[str] = None
+
+
+def _size(path: str) -> int:
+    try:
+        return os.path.getsize(path)
+    except OSError:
+        return 0
+
+
+def _free_bytes(path: str) -> int:
+    return shutil.disk_usage(path).free
+
+
+def _existing_ancestor(path: str) -> str:
+    current = os.path.abspath(path)
+    while not os.path.isdir(current):
+        parent = os.path.dirname(current)
+        if parent == current:
+            break
+        current = parent
+    return current
+
+
+def backup_names(db_path: str) -> Tuple[str, ...]:
+    """The pooled and pre-migration backups beside ``db_path`` (names)."""
+    folder = os.path.dirname(os.path.abspath(db_path))
+    base = os.path.basename(db_path)
+    prefixes = tuple(base + infix for infix in BACKUP_INFIXES)
+    try:
+        names = os.listdir(folder)
+    except OSError:
+        return ()
+    return tuple(
+        sorted(
+            name
+            for name in names
+            if name.startswith(prefixes) and os.path.isfile(os.path.join(folder, name))
+        )
+    )
+
+
+def plan_move(location) -> MovePlan:
+    """Check ``location`` can move to its target; raises MoveFailed."""
+    if not location.movable:
+        raise MoveFailed("not-movable")
+    target = os.path.abspath(location.target_dir)
+    source_dir = os.path.dirname(os.path.abspath(location.db_path))
+    if ll.same_folder(target, source_dir) or os.path.normcase(
+        os.path.realpath(target)
+    ) == os.path.normcase(os.path.realpath(source_dir)):
+        raise MoveFailed("same-folder")
+    if os.path.exists(target) and not os.path.isdir(target):
+        raise MoveFailed("target-invalid")
+    backups = backup_names(location.db_path)
+    # config.json moves with a node-folder library; one named by
+    # PROMPT_MANAGER_CONFIG_PATH is the user's and stays where it is.
+    config = None if location.config_explicit else location.config_path
+    config = config if config and os.path.isfile(config) else None
+    if ll.read_marker(target).get("state") != ll.MARKER_PENDING:
+        if os.path.exists(os.path.join(target, ll.DATABASE_NAME)):
+            raise MoveFailed("target-has-library")
+        names = [ll.CONFIG_NAME] if config else []
+        if any(os.path.exists(os.path.join(target, n)) for n in names + list(backups)):
+            raise MoveFailed("target-conflict")
+    db_bytes = _size(location.db_path) + _size(location.db_path + "-wal")
+    files_bytes = sum(_size(os.path.join(source_dir, n)) for n in backups)
+    files_bytes += _size(config) if config else 0
+    return MovePlan(
+        location,
+        os.path.abspath(location.db_path),
+        source_dir,
+        target,
+        config,
+        backups,
+        db_bytes,
+        files_bytes,
+    )
+
+
+def space_problem(plan: MovePlan) -> Optional[Tuple[int, int]]:
+    """``(required, free)`` when the target's volume is short, else None."""
+    free = _free_bytes(_existing_ancestor(plan.target_dir))
+    if free < plan.required_bytes:
+        return plan.required_bytes, free
+    return None
+
+
+# ── copy and verify ───────────────────────────────────────────────────
+
+
+def _tables(conn) -> List[str]:
+    return [
+        row[0]
+        for row in conn.execute(
+            "SELECT name FROM sqlite_master WHERE type='table' "
+            "AND name NOT LIKE 'sqlite_%' ORDER BY name"
+        )
+    ]
+
+
+def _describe(conn) -> Tuple[Dict[str, int], str]:
+    """Row counts per table and a digest of the workflow documents."""
+    counts = {}
+    for table in _tables(conn):
+        # A name read from sqlite_master, quoted as an identifier.
+        quoted = '"' + table.replace('"', '""') + '"'
+        sql = f"SELECT COUNT(*) FROM {quoted}"  # nosec B608
+        counts[table] = conn.execute(sql).fetchone()[0]
+    digest = hashlib.sha256()
+    if "workflow_documents" in counts:
+        for row in conn.execute(
+            "SELECT id, sha256, byte_length FROM workflow_documents ORDER BY id"
+        ):
+            digest.update(repr((row[0], bytes(row[1]), row[2])).encode())
+    return counts, digest.hexdigest()
+
+
+def _fsync_file(path: str) -> None:
+    with open(path, "rb+") as handle:
+        os.fsync(handle.fileno())
+
+
+def _fsync_dir(path: str) -> None:
+    if os.name != "posix":
+        return
+    fd = os.open(path, os.O_RDONLY)
+    try:
+        os.fsync(fd)
+    finally:
+        os.close(fd)
+
+
+def _private(path: str) -> None:
+    if os.name == "posix":
+        os.chmod(path, 0o600)
+
+
+def _copy_snapshot(source: sqlite3.Connection, temp: str):
+    """Back up ``source``'s current snapshot to ``temp``; return what it holds."""
+    source.execute("BEGIN")
+    try:
+        version = source.execute("PRAGMA data_version").fetchone()[0]
+        counts, documents = _describe(source)
+        destination = sqlite3.connect(temp)
+        try:
+            source.backup(destination)
+        finally:
+            destination.close()
+    finally:
+        source.execute("COMMIT")
+    with contextlib.closing(sqlite3.connect(temp)) as conn:
+        conn.execute("PRAGMA journal_mode=DELETE").fetchone()
+    _fsync_file(temp)
+    _private(temp)
+    return version, counts, documents
+
+
+def _verify_copy(temp: str, counts, documents) -> None:
+    ok, reason = PromptModel.verify_database_file(temp)
+    if not ok:
+        logger.error(f"Library move: the copy did not verify: {reason}")
+        detail = reason if reason.startswith("Workflow document") else None
+        raise MoveFailed(
+            "verify-failed",
+            f"{MESSAGES['verify-failed']} {detail}" if detail else None,
+        )
+    uri = Path(os.path.abspath(temp)).as_uri() + "?mode=ro"
+    try:
+        with contextlib.closing(sqlite3.connect(uri, uri=True)) as conn:
+            copied = _describe(conn)
+    except sqlite3.Error as error:
+        logger.error(f"Library move: the copy could not be read: {error}")
+        raise MoveFailed("verify-failed") from error
+    if copied != (counts, documents):
+        logger.error("Library move: the copy's rows or documents differ")
+        raise MoveFailed("verify-failed")
+
+
+def _hash_copy(source: str, destination: str) -> Tuple[int, str]:
+    """Copy ``source`` to ``destination`` (fsynced, owner-only, times kept)."""
+    digest = hashlib.sha256()
+    size = 0
+    with open(source, "rb") as reader, open(destination, "wb") as writer:
+        while True:
+            chunk = reader.read(_COPY_CHUNK)
+            if not chunk:
+                break
+            digest.update(chunk)
+            size += len(chunk)
+            writer.write(chunk)
+        writer.flush()
+        os.fsync(writer.fileno())
+    stat = os.stat(source)
+    os.utime(destination, ns=(stat.st_atime_ns, stat.st_mtime_ns))
+    _private(destination)
+    return size, digest.hexdigest()
+
+
+def _file_digest(path: str) -> Tuple[int, str]:
+    digest = hashlib.sha256()
+    size = 0
+    with open(path, "rb") as reader:
+        while True:
+            chunk = reader.read(_COPY_CHUNK)
+            if not chunk:
+                return size, digest.hexdigest()
+            digest.update(chunk)
+            size += len(chunk)
+
+
+def _check_copies(copies) -> None:
+    for source, temp, expected in copies:
+        if _file_digest(temp) != expected or _file_digest(source) != expected:
+            logger.error(
+                f"Library move: {os.path.basename(source)} did not copy unchanged"
+            )
+            raise MoveFailed("copy-failed")
+
+
+def write_marker(
+    directory: str, state: str, source_db: str, source_config: Optional[str] = None
+) -> str:
+    """Atomically write the move marker; return its path.
+
+    ``from`` and ``config_from`` name the old files, so a restart after a
+    crash between the switch and the rename can finish the rename.
+    """
+    path = os.path.join(directory, ll.MARKER_NAME)
+    temp = os.path.join(directory, f"{TEMP_PREFIX}{uuid.uuid4().hex}.marker")
+    payload = json.dumps(
+        {
+            "version": 1,
+            "state": state,
+            "from": source_db,
+            "config_from": source_config,
+            "updated_at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+        },
+        indent=2,
+    ).encode("utf-8")
+    fd = os.open(temp, os.O_WRONLY | os.O_CREAT | os.O_EXCL, 0o600)
+    try:
+        with os.fdopen(fd, "wb") as handle:
+            handle.write(payload)
+            handle.flush()
+            os.fsync(handle.fileno())
+        _private(temp)
+        os.replace(temp, path)
+    except BaseException:
+        _remove(temp)
+        raise
+    _fsync_dir(directory)
+    return path
+
+
+def _remove(*paths) -> None:
+    for path in paths:
+        try:
+            if path and os.path.lexists(path):
+                os.remove(path)
+        except OSError:
+            logger.warning(f"Library move: could not remove {os.path.basename(path)}")
+
+
+def sweep_move_temps(directory: str) -> List[str]:
+    """Remove the temp files an interrupted move left in ``directory``."""
+    removed = []
+    try:
+        names = os.listdir(directory)
+    except OSError:
+        return removed
+    for name in names:
+        if name.startswith(TEMP_PREFIX):
+            _remove(os.path.join(directory, name))
+            removed.append(name)
+    return removed
+
+
+# ── the move ─────────────────────────────────────────────────────────
+
+
+def _noop(*_args, **_kwargs):
+    return None
+
+
+def switch_process(old_db: str, location, detached: list) -> None:
+    """Point this process at ``location``: resolver, every model, observers.
+
+    Adds the old file's connections, detached and still open, to
+    ``detached`` as soon as they are detached, so the mover retires them
+    (``PromptModel.retire_detached``) whether or not the rest succeeds.
+    """
+    from .execution_history import rekey_history_restore
+    from .review_changes import rekey_review_changes
+
+    ll.switch_to(location)
+    detached.extend(PromptModel.repoint_instances(old_db, location.db_path))
+    rekey_review_changes(old_db, location.db_path)
+    rekey_history_restore(old_db, location.db_path)
+
+
+def unswitch_process(old_db: str, new_db: str) -> list:
+    """Undo ``switch_process`` (any part of it); returns detached connections."""
+    from .execution_history import rekey_history_restore
+    from .review_changes import rekey_review_changes
+
+    ll.clear_override()
+    detached = PromptModel.repoint_instances(new_db, old_db)
+    rekey_review_changes(new_db, old_db)
+    rekey_history_restore(new_db, old_db)
+    return detached
+
+
+class _Move:
+    def __init__(self, plan, *, switch, config_lock, progress, checkpoint):
+        self.plan = plan
+        self.switch = switch
+        self.config_lock = config_lock
+        self.progress = progress
+        self.checkpoint = checkpoint
+        self.created_dir = False
+        self.temps: List[str] = []
+        self.published: List[str] = []
+        self.marker: Optional[str] = None
+        self.moved = False
+        # Retired connections still in use; awaited once the locks are free.
+        self.retiring: list = []
+
+    def temp(self, suffix: str) -> str:
+        path = os.path.join(
+            self.plan.target_dir, f"{TEMP_PREFIX}{uuid.uuid4().hex}{suffix}"
+        )
+        self.temps.append(path)
+        return path
+
+    def prepare_target(self) -> None:
+        target = self.plan.target_dir
+        self.created_dir = not os.path.isdir(target)
+        ll.ensure_private_dir(target)
+        sweep_move_temps(target)
+        if ll.read_marker(target).get("state") == ll.MARKER_PENDING:
+            # An earlier move stopped before it switched: its files are ours.
+            for name in (ll.DATABASE_NAME, ll.CONFIG_NAME) + self.plan.backups:
+                _remove(os.path.join(target, name))
+            for suffix in ("-wal", "-shm", "-journal"):
+                _remove(os.path.join(target, ll.DATABASE_NAME + suffix))
+
+    def copy_round(self, source):
+        """Steps 2–4 once: the verified temp copies and the snapshot version."""
+        self.progress("Copying the database")
+        temp_db = self.temp(".db")
+        version, counts, documents = _copy_snapshot(source, temp_db)
+        self.checkpoint("copied")
+        self.progress("Verifying the copy")
+        _verify_copy(temp_db, counts, documents)
+        self.checkpoint("verified")
+        copies = []
+        for name in self.plan.backups:
+            src = os.path.join(self.plan.source_dir, name)
+            temp = self.temp(".bak")
+            try:
+                expected = _hash_copy(src, temp)
+            except OSError as error:
+                logger.error(f"Library move: copying {name} failed: {error}")
+                raise MoveFailed("copy-failed") from error
+            copies.append((src, temp, expected, name))
+        if copies:
+            self.progress(f"Copied {len(copies)} backup file(s)")
+        self.checkpoint("files")
+        _check_copies([(s, t, e) for s, t, e, _ in copies])
+        return version, counts, temp_db, copies
+
+    def discard_round(self, temp_db, copies) -> None:
+        _remove(temp_db, *[temp for _s, temp, _e, _n in copies])
+
+    def publish(self, temp_db, copies) -> List[str]:
+        """Under the locks: config.json, marker, final names, marker moved."""
+        target = self.plan.target_dir
+        files = []
+        if self.plan.config_path:
+            temp = self.temp(".config")
+            try:
+                expected = _hash_copy(self.plan.config_path, temp)
+            except OSError as error:
+                raise MoveFailed("copy-failed") from error
+            _check_copies([(self.plan.config_path, temp, expected)])
+            copies = copies + [(self.plan.config_path, temp, expected, ll.CONFIG_NAME)]
+        self.marker = write_marker(
+            target, ll.MARKER_PENDING, self.plan.db_path, self.plan.config_path
+        )
+        self.checkpoint("pending")
+        for _src, temp, _expected, name in copies:
+            final = os.path.join(target, name)
+            os.replace(temp, final)
+            self.published.append(final)
+            files.append(name)
+        final_db = os.path.join(target, ll.DATABASE_NAME)
+        os.replace(temp_db, final_db)
+        self.published.append(final_db)
+        files.append(ll.DATABASE_NAME)
+        _fsync_dir(target)
+        self.checkpoint("published")
+        write_marker(target, ll.MARKER_MOVED, self.plan.db_path, self.plan.config_path)
+        self.moved = True
+        self.checkpoint("marked")
+        return files
+
+    def cleanup(self) -> None:
+        if self.moved:
+            return
+        _remove(*self.temps, *self.published, self.marker)
+        if self.created_dir:
+            try:
+                os.rmdir(self.plan.target_dir)
+            except OSError:
+                pass
+
+
+def _unique(path: str) -> str:
+    candidate, number = path, 1
+    while os.path.lexists(candidate):
+        candidate = f"{path}_{number}"
+        number += 1
+    return candidate
+
+
+def retire_old_files(plan: MovePlan) -> Tuple[List[str], Optional[str]]:
+    """Rename the old database and config.json; ``(renamed, problem code)``.
+
+    Every connection to the old file must be closed. The WAL is checkpointed
+    first; when it cannot be emptied (another process still has the file
+    open) nothing is renamed (``old-in-use``).
+    """
+    old = plan.db_path
+    folder = os.path.basename(os.path.normpath(plan.target_dir))
+    try:
+        with contextlib.closing(sqlite3.connect(old, timeout=1)) as conn:
+            busy = conn.execute("PRAGMA wal_checkpoint(TRUNCATE)").fetchone()[0]
+    except sqlite3.Error as error:
+        logger.warning(
+            f"Library move: the old database could not be checkpointed: {error}"
+        )
+        return [], "old-in-use"
+    if busy or _size(old + "-wal") > 0:
+        return [], "old-in-use"
+    renamed = []
+    try:
+        for path in (old, plan.config_path):
+            if not path or not os.path.isfile(path):
+                continue
+            new_name = _unique(path + MOVED_INFIX + folder)
+            os.rename(path, new_name)
+            renamed.append(os.path.basename(new_name))
+        for suffix in ("-wal", "-shm"):
+            if os.path.isfile(old + suffix) and _size(old + suffix) == 0:
+                os.remove(old + suffix)
+    except OSError as error:
+        logger.warning(f"Library move: the old files could not be renamed: {error}")
+        return renamed, "old-locked"
+    return renamed, None
+
+
+def finish_interrupted_move(location) -> List[str]:
+    """Rename old files a crash left between the switch and the rename.
+
+    The marker of the folder in use says ``moved`` and names the old files;
+    an old ``prompts.db`` (and config.json) still there under its own name is
+    renamed ``*.moved-to-<folder>`` now, never deleted. Returns the new
+    names (nothing when there is nothing to finish or the file is in use).
+    """
+    if location.source not in (ll.SOURCE_DEFAULT, ll.SOURCE_ENVIRONMENT):
+        return []
+    marker = ll.read_marker(location.data_dir)
+    old = marker.get("from")
+    if marker.get("state") != ll.MARKER_MOVED or not isinstance(old, str):
+        return []
+    if not os.path.isfile(old) or os.path.basename(old) != ll.DATABASE_NAME:
+        return []
+    if ll.same_folder(os.path.dirname(old), location.data_dir):
+        return []
+    config = marker.get("config_from")
+    if not (
+        isinstance(config, str)
+        and os.path.basename(config) == ll.CONFIG_NAME
+        and os.path.isfile(config)
+        and os.path.normcase(os.path.abspath(config))
+        != os.path.normcase(os.path.abspath(location.config_path))
+    ):
+        config = None
+    pending = types.SimpleNamespace(
+        db_path=old, config_path=config, target_dir=location.data_dir
+    )
+    renamed, problem = retire_old_files(pending)
+    if problem:
+        logger.warning(f"Library move: the old files are still in use ({problem})")
+    elif renamed:
+        logger.info(f"Library move finished: renamed {', '.join(renamed)}")
+    return renamed
+
+
+def move_library(
+    plan: MovePlan,
+    *,
+    switch: Callable = switch_process,
+    unswitch: Callable = unswitch_process,
+    after_switch: Callable = _noop,
+    config_lock=None,
+    progress: Callable[[str], None] = _noop,
+    checkpoint: Callable[[str], None] = _noop,
+    attempts: int = DEFAULT_ATTEMPTS,
+) -> MoveResult:
+    """Move the library as planned (blocking; see the module docstring).
+
+    ``switch(old_db, location, detached)`` points the process at the copy
+    and adds the old connections it detached to ``detached``; if it raises,
+    ``unswitch(old_db, new_db)`` undoes it while both files are still
+    locked, every detached connection is closed, the copy is removed and
+    the move fails (``switch-failed``). ``after_switch()`` runs
+    once the switch is final; it is best effort (logged, never fails the move).
+    """
+    short = space_problem(plan)
+    if short is not None:
+        raise MoveFailed("insufficient-space")
+    lock = config_lock if config_lock is not None else contextlib.nullcontext()
+    move = _Move(
+        plan,
+        switch=switch,
+        config_lock=lock,
+        progress=progress,
+        checkpoint=checkpoint,
+    )
+    move.unswitch = unswitch
+    location = plan.target_location
+    try:
+        move.prepare_target()
+        files, counts = _copy_and_switch(move, location, attempts)
+    except Exception:
+        move.cleanup()
+        raise
+    try:
+        after_switch()
+    except Exception:
+        logger.exception("Library move: a follow-up after the switch failed")
+    from .execution_history import notify_history_restore
+
+    notify_history_restore(location.db_path)
+    progress("Renaming the old files")
+    renamed, problem = retire_old_files(plan)
+    checkpoint("retired")
+    return MoveResult(location, counts, files, renamed, problem)
+
+
+def _switch_guarded(move: _Move, location) -> list:
+    """Switch while the copy is locked too; undo it all if any part fails.
+
+    The copy's write lock keeps every repointed thread from writing to it
+    until the switch is whole, so undoing a partial switch loses nothing.
+    Returns the old file's detached connections (still open). On failure
+    every detached connection is retired; those still in use are added to
+    ``move.retiring`` for the caller to await once the locks are released.
+    """
+    plan = move.plan
+    guard = sqlite3.connect(
+        location.db_path, timeout=BUSY_TIMEOUT_SECONDS, isolation_level=None
+    )
+    try:
+        guard.execute("BEGIN IMMEDIATE")
+        detached: list = []
+        try:
+            move.switch(plan.db_path, location, detached)
+        except Exception as error:
+            logger.exception("Library move: switching failed; undoing it")
+            move.moved = False  # the copy and its marker are removed
+            # Retire, never wait, here: a connection still in use may be
+            # waiting on the very locks this thread holds.
+            try:
+                undone = move.unswitch(plan.db_path, location.db_path)
+            except Exception:
+                logger.exception("Library move: undoing the switch failed")
+            else:
+                _retire(move, undone)
+            # The old file's connections a partial switch detached: the
+            # models reopen it on next use, so close them now, not at GC.
+            _retire(move, detached)
+            raise MoveFailed("switch-failed") from error
+        move.checkpoint("switched")
+        return detached
+    finally:
+        if guard.in_transaction:
+            guard.execute("ROLLBACK")
+        guard.close()
+
+
+def _copy_and_switch(move: _Move, location, attempts: int):
+    """Copy, then switch under both locks; retired connections are awaited.
+
+    A connection that was inside a statement when it was retired is closed
+    by its own thread once the statement ends, which may need the locks
+    released first; the old files are renamed (and a failed copy removed)
+    only after that, since Windows refuses to rename or delete an open file.
+    """
+    try:
+        return _copy_and_switch_locked(move, location, attempts)
+    finally:
+        try:
+            PromptModel.await_detached(move.retiring, logger)
+        except Exception:
+            logger.exception("Library move: waiting for old connections failed")
+
+
+def _retire(move: _Move, connections: list) -> None:
+    """Retire ``connections``, adding those still in use to ``move.retiring``.
+
+    Never raises: a switch that completed must not be reported as failed
+    because one of the old connections could not be retired (logged; the
+    rename then reports the old files as still in use).
+    """
+    try:
+        move.retiring.extend(PromptModel.retire_detached(connections, logger))
+    except Exception:
+        logger.exception("Library move: retiring the old connections failed")
+
+
+def _copy_and_switch_locked(move: _Move, location, attempts: int):
+    plan = move.plan
+    for attempt in range(1, attempts + 1):
+        source = sqlite3.connect(
+            plan.db_path, timeout=BUSY_TIMEOUT_SECONDS, isolation_level=None
+        )
+        try:
+            source.execute(f"PRAGMA busy_timeout = {BUSY_TIMEOUT_SECONDS * 1000}")
+            version, counts, temp_db, copies = move.copy_round(source)
+            source.execute("BEGIN IMMEDIATE")
+            try:
+                if source.execute("PRAGMA data_version").fetchone()[0] != version:
+                    move.progress("The library changed during the copy; copying again")
+                    move.discard_round(temp_db, copies)
+                    continue
+                move.checkpoint("locked")
+                with move.config_lock:
+                    files = move.publish(temp_db, copies)
+                    detached = _switch_guarded(move, location)
+                # Still holding the old file's lock: a write that was waiting
+                # on it is interrupted and cannot commit, so it fails (and
+                # runs again on the new library) instead of landing in the
+                # old file. Its own thread closes the connection.
+                _retire(move, detached)
+            finally:
+                if source.in_transaction:
+                    source.execute("ROLLBACK")
+            return files, counts
+        finally:
+            source.close()
+    raise MoveFailed("library-busy")

database/library_schema.py

@@ -0,0 +1,86 @@
+"""Optional-old, required-current schema for Library parent state."""
+
+import re
+import sqlite3
+
+TABLE = "prompt_library_state"
+MAX_TITLE_LENGTH = 10_000
+_DDL = (
+    f"CREATE TABLE {TABLE} (prompt_id INTEGER PRIMARY KEY, title TEXT, "
+    "FOREIGN KEY(prompt_id) REFERENCES prompts(id) ON DELETE CASCADE)"
+)
+_COLUMNS = (("prompt_id", "INTEGER", 0, 1), ("title", "TEXT", 0, 0))
+_FOREIGN_KEYS = {("prompt_id", "prompts", "id", "CASCADE")}
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql).lower()
+
+
+def library_schema_error(
+    conn: sqlite3.Connection, *, allow_missing: bool = True
+) -> str | None:
+    """Reject unsupported Library state schemas and invalid persisted values.
+
+    Reads only: nothing is ever repaired. A missing table is acceptable for
+    backups that predate it unless ``allow_missing`` is false.
+    """
+    objects = conn.execute(
+        "SELECT type,sql FROM sqlite_master WHERE name=? "
+        "OR (type IN ('index','trigger') AND tbl_name=?)",
+        (TABLE, TABLE),
+    ).fetchall()
+    if not objects:
+        return None if allow_missing else "Missing Library state schema"
+    tables = [sql for kind, sql in objects if kind == "table"]
+    if len(tables) != 1 or len(objects) != 1:
+        return "Library state name is reserved by an unsupported schema object"
+    if _compact(tables[0] or "") != _compact(_DDL):
+        return "Library state has an invalid table definition"
+    columns = tuple(
+        (row[1], row[2].upper(), row[3], row[5])
+        for row in conn.execute(f"PRAGMA table_info({TABLE})")
+    )
+    if columns != _COLUMNS:
+        return "Library state has invalid columns"
+    foreign_keys = {
+        (row[3], row[2], row[4], row[6])
+        for row in conn.execute(f"PRAGMA foreign_key_list({TABLE})")
+    }
+    if foreign_keys != _FOREIGN_KEYS:
+        return "Library state has invalid foreign key definitions"
+    if conn.execute(f"PRAGMA foreign_key_check({TABLE})").fetchone():
+        return "Library state contains foreign key violations"
+    # length() stops at an embedded NUL, so bound by bytes in SQL (characters
+    # never outnumber bytes, and UTF-8 uses at most 4 bytes per character) and
+    # count the few remaining candidates as decoded characters.
+    if conn.execute(
+        f"SELECT 1 FROM {TABLE} WHERE typeof(title) NOT IN ('null','text') "
+        f"OR length(CAST(title AS BLOB)) > {MAX_TITLE_LENGTH * 4} LIMIT 1"
+    ).fetchone():
+        return "Library state contains invalid titles"
+    for (raw,) in conn.execute(
+        f"SELECT CAST(title AS BLOB) FROM {TABLE} "
+        f"WHERE length(CAST(title AS BLOB)) > {MAX_TITLE_LENGTH}"
+    ):
+        if len(bytes(raw).decode("utf-8", "replace")) > MAX_TITLE_LENGTH:
+            return "Library state contains invalid titles"
+    return None
+
+
+def migrate_library_state(conn: sqlite3.Connection) -> None:
+    """Install the table atomically; validate before and after creating it."""
+    conn.execute("SAVEPOINT prompt_library_state_schema")
+    try:
+        error = library_schema_error(conn)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute(_DDL.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        error = library_schema_error(conn, allow_missing=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute("RELEASE prompt_library_state_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO prompt_library_state_schema")
+        conn.execute("RELEASE prompt_library_state_schema")
+        raise

database/library_tokens.py

@@ -0,0 +1,37 @@
+"""Narrow comparison tokens for a Library parent's own metadata projection."""
+
+import hashlib
+import json
+
+_DOMAIN = "prompt-manager-library-parent:v1\n"
+
+
+def canonical_token(domain: str, preimage: dict) -> str:
+    """SHA-256 of ``domain`` plus the canonical JSON of ``preimage``."""
+    payload = json.dumps(
+        preimage,
+        sort_keys=True,
+        separators=(",", ":"),
+        ensure_ascii=False,
+        allow_nan=False,
+    )
+    return hashlib.sha256((domain + payload).encode("utf-8")).hexdigest()
+
+
+def library_edit_token(epoch: str, positive: dict) -> str:
+    """Hash the parent's owner fields under the current storage epoch.
+
+    Timestamps, counts, pairs and previews are excluded, and an absent state
+    row projects the same null title as a default one.
+    """
+    preimage = {
+        "epoch": epoch,
+        "id": positive["id"],
+        "text": positive["text"],
+        "title": positive["title"],
+        "category": positive["category"],
+        "notes": positive["notes"],
+        "rating": positive["rating"],
+        "tags": sorted(set(positive["tags"])),
+    }
+    return canonical_token(_DOMAIN, preimage)

database/maintenance.py

@@ -0,0 +1,740 @@
+"""Reviewed maintenance plans: read-only previews and their exact apply (#249 S6).
+
+Four operations, each the native owner of a legacy ``/maintenance`` one and
+built on the same queries, so a preview counts what the legacy cleanup would
+change:
+
+- ``duplicate_prompts``: prompts whose text is equal ignoring case and outer
+  spaces (``DUPLICATE_PROMPTS_SQL``, legacy ``cleanup_duplicates``); each
+  group keeps its oldest prompt, which takes over the others' images and
+  metadata. A group with a member in a saved pair is kept whole.
+- ``optimize``: VACUUM (legacy ``vacuum``).
+- ``unused_components``: prompts nothing references (``ORPHANED_PROMPTS_SQL``,
+  legacy ``prune_orphaned_prompts``, with the #239 protections).
+- ``missing_links``: image rows whose file is gone (legacy
+  ``cleanup_orphaned_images``), counted only under a source that can be read
+  now, as the health check's ``missing_files`` judges them
+  (``health.link_states``). It deletes database rows only, never a file.
+
+``preview_maintenance`` reads one snapshot on its own ``mode=ro`` connection
+(``health.readonly_uri``) and closes it in ``finally``; files are looked at
+after the snapshot ended. It returns a ``MaintenanceDraft``: the public plan
+fields, the targets an apply needs and the fingerprint of exactly what was
+reviewed (ids, text hashes, metadata, tags, Library state and pair refs of
+every prompt involved, or each link's row and source). The fingerprint never
+leaves the server.
+
+``apply_maintenance`` recomputes that fingerprint on the write connection
+inside ``BEGIN IMMEDIATE`` and raises ``PlanChanged`` before writing anything
+when it differs; otherwise it applies the reviewed targets in batches of
+``ID_BATCH`` in that one transaction. VACUUM cannot run in a transaction, so
+``optimize`` binds to its database only (the caller checks that).
+"""
+
+import hashlib
+import json
+import os
+import shutil
+import sqlite3
+from dataclasses import dataclass, field
+from pathlib import Path
+
+from .execution_contexts import LIVE_HINT_CTES
+from .health import BATCH_ROWS, link_states, readable_roots, readonly_uri
+from .pairs import PairDependencyError
+from .workflow_documents import documents_of_rows, release_documents
+
+MAINTENANCE_OPERATIONS = (
+    "duplicate_prompts",
+    "optimize",
+    "unused_components",
+    "missing_links",
+    # Applied as its own batched job (database/workflow_storage.py).
+    "workflow_storage",
+)
+MAX_DETAILS = 50  # detail rows a plan lists
+MAX_RESULT_ROWS = 200  # changed records an apply describes (the job lists 200)
+EXCERPT_CHARS = 80
+# Ids bound per statement: the pair-reference read binds each id twice, which
+# stays under SQLite's historical 999-variable limit.
+ID_BATCH = 400
+
+# Why a prompt no image links to is kept (#239): predicates on ``p``, in the
+# order the plan names them. ORPHANED_PROMPTS_SQL is built from them, so the
+# reasons a plan lists are exactly the rules the prune applies.
+PROTECTIONS = (
+    (
+        "EXISTS (SELECT 1 FROM prompt_pairs pp "
+        "WHERE pp.positive_prompt_id = p.id OR pp.negative_prompt_id = p.id)",
+        "in saved pairs",
+        "in a saved pair",
+    ),
+    (
+        "EXISTS (SELECT 1 FROM prompt_tags pt JOIN tags t ON pt.tag_id = t.id "
+        "WHERE pt.prompt_id = p.id AND t.name = '__protected__')",
+        "tagged __protected__",
+        "tagged __protected__",
+    ),
+    (
+        "EXISTS (SELECT 1 FROM prompt_library_state ls WHERE ls.prompt_id = p.id)",
+        "with Library state (a title or an explicit parent)",
+        "with Library state (a title or an explicit parent)",
+    ),
+    (
+        "p.id IN (SELECT prompt_id FROM live_hint_prompts)",
+        "used by recorded runs",
+        "used by a recorded run",
+    ),
+)
+
+# Exempt references: images, pairs, the __protected__ tag, Library parent
+# state, and live recorded-run hints (computed once, see LIVE_HINT_CTES).
+ORPHANED_PROMPTS_SQL = (
+    "WITH "
+    + LIVE_HINT_CTES
+    + " SELECT p.id FROM prompts p"
+    + " LEFT JOIN generated_images gi ON p.id = gi.prompt_id"
+    + " WHERE gi.prompt_id IS NULL"
+    + "".join(f" AND NOT ({predicate})" for predicate, _many, _one in PROTECTIONS)
+)
+# The prompts no image links to, with which protections keep each one.
+_UNLINKED_PROTECTIONS_SQL = (
+    "WITH "
+    + LIVE_HINT_CTES
+    + " SELECT "
+    + ", ".join(
+        f"COALESCE(SUM({predicate}), 0)" for predicate, _many, _one in PROTECTIONS
+    )
+    + ", COALESCE(SUM("
+    + " OR ".join(f"({predicate})" for predicate, _many, _one in PROTECTIONS)
+    + "), 0) FROM prompts p WHERE NOT EXISTS "
+    "(SELECT 1 FROM generated_images gi WHERE gi.prompt_id = p.id)"
+)
+
+# Duplicate prompt groups, as the legacy cleanup finds them (case-insensitive
+# text; GROUP_CONCAT order is sorted by created_at afterwards).
+DUPLICATE_PROMPTS_SQL = """
+    SELECT LOWER(TRIM(text)) as normalized_text, COUNT(*) as count,
+           GROUP_CONCAT(id) as ids,
+           GROUP_CONCAT(created_at) as created_dates
+    FROM prompts
+    GROUP BY LOWER(TRIM(text))
+    HAVING COUNT(*) > 1
+"""
+
+_LINK_ROWS_SQL = (
+    "SELECT id, image_path, filename, file_path FROM generated_images ORDER BY id"
+)
+
+# What a prompt's fingerprint covers, per batch of ids.
+_PROMPT_STATE_SQL = (
+    "SELECT id, hash, text, category, rating, notes, created_at, updated_at "
+    "FROM prompts WHERE id IN ({marks}) ORDER BY id",
+    "SELECT prompt_id, tag_id FROM prompt_tags WHERE prompt_id IN ({marks}) "
+    "ORDER BY prompt_id, tag_id",
+    "SELECT * FROM prompt_library_state WHERE prompt_id IN ({marks}) "
+    "ORDER BY prompt_id",
+    "SELECT id, positive_prompt_id, negative_prompt_id FROM prompt_pairs "
+    "WHERE positive_prompt_id IN ({marks}) OR negative_prompt_id IN ({marks}) "
+    "ORDER BY id",
+)
+
+NO_FILES_DELETED = "No image file is deleted."
+LINKS_ONLY = "Database links only; no image file is deleted."
+UNUSED_DETAIL = "No image, saved pair, Library state or recorded run uses it"
+
+
+# Why a duplicate group is kept whole (``_duplicate_conflict``).
+CONFLICT_REASONS = {
+    "collision": ("group links one image twice", "groups link one image twice"),
+    "title": (
+        "group has its own Library title",
+        "groups have their own Library titles",
+    ),
+}
+CONFLICT_DETAILS = {
+    "collision": "Kept: two copies link the same image",
+    "title": "Kept: a copy has its own Library title",
+}
+OPTIMIZE_LOCK_NOTE = (
+    "Runs alone: the database is locked while it runs, and on a large library "
+    "this takes a while. Prompt runs counted and images captured meanwhile "
+    "wait up to 30 seconds each; anything that waits longer is not recorded."
+)
+
+
+class PlanChanged(Exception):
+    """What a plan was reviewed on changed before it was applied."""
+
+
+class MergeConflict(Exception):
+    """A strict duplicate merge would drop an image link or a Library title."""
+
+
+@dataclass
+class MaintenanceDraft:
+    """A preview: the public plan fields plus what only the server keeps."""
+
+    operation: str
+    fingerprint: str
+    targets: list
+    scope: dict
+    protected: dict
+    media: dict
+    details: list = field(default_factory=list)
+
+    def public(self):
+        """The plan fields a response may carry (no fingerprint, no targets)."""
+        return {
+            "operation": self.operation,
+            "scope": dict(self.scope),
+            "protected": {
+                "count": self.protected["count"],
+                "reasons": list(self.protected["reasons"]),
+            },
+            "media": dict(self.media),
+            "details": [dict(row) for row in self.details],
+        }
+
+
+def _plural(count, one, many=None):
+    return f"{count:,} {one if count == 1 else (many or one + 's')}"
+
+
+def _excerpt(text):
+    text = " ".join(str(text or "").split())
+    if len(text) <= EXCERPT_CHARS:
+        return text
+    return text[: EXCERPT_CHARS - 1].rstrip() + "…"
+
+
+def _megabytes(size):
+    return f"{size / 2**20:,.1f} MB"
+
+
+class _Fingerprint:
+    """A streaming sha256 over JSON-encoded parts."""
+
+    def __init__(self, operation):
+        self._hash = hashlib.sha256()
+        self.add(operation)
+
+    def add(self, part):
+        encoded = json.dumps(part, separators=(",", ":"), default=str)
+        self._hash.update(encoded.encode("utf-8") + b"\n")
+
+    def hexdigest(self):
+        return "sha256:" + self._hash.hexdigest()
+
+
+def _batches(ids):
+    ids = list(ids)
+    for start in range(0, len(ids), ID_BATCH):
+        yield ids[start : start + ID_BATCH]
+
+
+def _marks(batch):
+    return ",".join("?" for _ in batch)
+
+
+def _add_prompt_state(fingerprint, conn, ids):
+    """Fold every reviewed fact about the prompts ``ids`` into ``fingerprint``."""
+    for batch in _batches(sorted(ids)):
+        marks = _marks(batch)
+        for sql in _PROMPT_STATE_SQL:
+            params = batch * sql.count("{marks}")
+            for row in conn.execute(sql.format(marks=marks), params):
+                fingerprint.add(list(row))
+
+
+def _texts(conn, ids, limit=MAX_DETAILS):
+    """``{id: text}`` of the first ``limit`` prompts, for detail rows."""
+    ids = list(ids)[:limit]
+    if not ids:
+        return {}
+    rows = conn.execute(
+        f"SELECT id, text FROM prompts WHERE id IN ({_marks(ids)})", ids
+    )
+    return {row[0]: row[1] for row in rows}
+
+
+def _orphaned_ids(conn):
+    return sorted(row[0] for row in conn.execute(ORPHANED_PROMPTS_SQL))
+
+
+def _delete_ids(conn, table, ids):
+    """Delete ``table`` rows by id in batches; returns the rows removed.
+
+    The workflow documents only those rows referenced go with them (a
+    prompt's images go by cascade), in the same transaction.
+    """
+    if table not in ("prompts", "generated_images"):
+        raise ValueError(f"Unsupported table {table!r}")
+    removed = 0
+    for batch in _batches(ids):
+        documents = documents_of_rows(
+            conn, "id" if table == "generated_images" else "prompt_id", batch
+        )
+        cursor = conn.execute(
+            f"DELETE FROM {table} WHERE id IN ({_marks(batch)})",  # nosec B608
+            batch,
+        )
+        removed += cursor.rowcount
+        release_documents(conn, documents)
+    return removed
+
+
+def _read_snapshot(db_path, read):
+    """``read(conn)`` on one ``mode=ro`` snapshot; the connection always closes."""
+    conn = sqlite3.connect(readonly_uri(db_path), uri=True)
+    try:
+        conn.row_factory = sqlite3.Row
+        conn.execute("BEGIN")  # one read snapshot for the whole plan
+        return read(conn)
+    finally:
+        conn.close()
+
+
+def _relative(image_path, root):
+    try:
+        return Path(os.path.relpath(image_path, root)).as_posix()
+    except ValueError:  # another drive (Windows)
+        return os.path.basename(image_path)
+
+
+def _source_name(root, index):
+    return os.path.basename(str(root).rstrip("\\/")) or f"Source {index + 1}"
+
+
+class MaintenanceOperations:
+    """Mixin for PromptDatabase: maintenance previews and their apply."""
+
+    # ── shared with the legacy cleanup ───────────────────────────────
+
+    def _duplicate_groups(self, conn):
+        """``(groups, paired)``: duplicate id groups, oldest first.
+
+        ``paired`` are the groups with a member in a saved pair, which the
+        cleanup keeps whole.
+        """
+        groups, paired = [], []
+        for duplicate in conn.execute(DUPLICATE_PROMPTS_SQL).fetchall():
+            ids = duplicate["ids"].split(",")
+            created_dates = duplicate["created_dates"].split(",")
+            # Sort IDs by created_at date to keep the oldest
+            id_date_pairs = list(zip(ids, created_dates))
+            id_date_pairs.sort(key=lambda x: x[1])
+            sorted_ids = [int(pair[0]) for pair in id_date_pairs]
+            try:
+                self._ensure_prompts_unpaired(conn, sorted_ids)
+            except PairDependencyError:
+                paired.append(sorted_ids)
+                continue
+            groups.append(sorted_ids)
+        return groups, paired
+
+    @staticmethod
+    def _duplicate_conflict(conn, sorted_ids):
+        """Why merging the group would lose something, or None.
+
+        ``collision``: two members link the same file, so moving the images
+        to the oldest prompt would hit ``UNIQUE(prompt_id, file_path)``.
+        ``title``: a member that would be deleted has its own Library title.
+        """
+        marks = _marks(sorted_ids)
+        collision = conn.execute(
+            "SELECT 1 FROM generated_images WHERE prompt_id IN "  # nosec B608
+            f"({marks}) AND file_path IS NOT NULL GROUP BY file_path "
+            "HAVING COUNT(DISTINCT prompt_id) > 1 LIMIT 1",
+            sorted_ids,
+        ).fetchone()
+        if collision:
+            return "collision"
+        duplicates = sorted_ids[1:]
+        titled = conn.execute(
+            "SELECT 1 FROM prompt_library_state WHERE prompt_id IN "  # nosec B608
+            f"({_marks(duplicates)}) AND trim(coalesce(title, '')) != '' LIMIT 1",
+            duplicates,
+        ).fetchone()
+        return "title" if titled else None
+
+    def _merge_duplicate_group(self, conn, sorted_ids, strict=False):
+        """Keep the oldest prompt of a group; returns (removed, images moved).
+
+        ``strict`` (the reviewed apply): a merge that would drop an image
+        link or a Library title raises MergeConflict, and every helper error
+        propagates, so the transaction rolls back. The legacy cleanup keeps
+        its old, lenient behaviour.
+        """
+        primary_id, duplicate_ids = sorted_ids[0], sorted_ids[1:]
+        if strict:
+            conflict = self._duplicate_conflict(conn, sorted_ids)
+            if conflict:
+                raise MergeConflict(conflict)
+        self.logger.debug(
+            f"Merging duplicates: keeping {primary_id}, removing {duplicate_ids}"
+        )
+        primary = conn.execute(
+            "SELECT * FROM prompts WHERE id = ?", (primary_id,)
+        ).fetchone()
+        if not primary:
+            return 0, 0
+        merged_metadata = self._merge_duplicate_metadata(
+            conn, primary_id, duplicate_ids, strict=strict
+        )
+        moved = self._transfer_images_to_primary(
+            conn, primary_id, duplicate_ids, strict=strict
+        )
+        if merged_metadata:
+            self._update_primary_with_merged_metadata(
+                conn, primary_id, merged_metadata, strict=strict
+            )
+        removed = 0
+        # Images the lenient merge could not move go with their prompt.
+        documents = documents_of_rows(conn, "prompt_id", duplicate_ids)
+        for duplicate_id in duplicate_ids:
+            conn.execute("DELETE FROM prompts WHERE id = ?", (duplicate_id,))
+            removed += 1
+            self.logger.debug(f"Removed duplicate prompt {duplicate_id}")
+        release_documents(conn, documents)
+        return removed, moved
+
+    @staticmethod
+    def _missing_link_ids(rows, roots):
+        """Ids of the ``rows`` whose file is gone under a readable source."""
+        readable = readable_roots(roots)
+        return [
+            row[0]
+            for row, _owner, state in link_states(rows, roots, readable)
+            if state == "missing"
+        ]
+
+    # ── previews ──────────────────────────────────────────────────────
+
+    def preview_maintenance(self, operation, roots=()) -> MaintenanceDraft:
+        """The reviewed plan of ``operation``; writes nothing (blocking).
+
+        ``roots`` are the sources missing links are judged under (the health
+        check's: the effective sources plus configured ones that are gone).
+        """
+        if operation not in MAINTENANCE_OPERATIONS:
+            raise ValueError(f"Unknown maintenance operation {operation!r}")
+        roots = [str(root) for root in roots]
+        if operation == "optimize":
+            return self._plan_optimize()
+        if operation == "workflow_storage":
+            return self._plan_workflow_storage()
+        if operation == "missing_links":
+            rows = _read_snapshot(self.model.db_path, self._link_rows)
+            return self._plan_missing_links(rows, roots)
+        reader = (
+            self._plan_duplicate_prompts
+            if operation == "duplicate_prompts"
+            else self._plan_unused_components
+        )
+        return _read_snapshot(self.model.db_path, reader)
+
+    @staticmethod
+    def _link_rows(conn):
+        rows = []
+        cursor = conn.execute(_LINK_ROWS_SQL)
+        while batch := cursor.fetchmany(BATCH_ROWS):
+            rows.extend(tuple(row) for row in batch)
+        return rows
+
+    def _duplicates_fingerprint(self, conn):
+        """``(groups, paired, conflicted, fingerprint)`` of the duplicate merge.
+
+        ``conflicted`` holds ``(ids, reason)`` of the groups a merge would
+        lose an image link or a Library title in (``_duplicate_conflict``);
+        the plan protects them.
+        """
+        candidates, paired = self._duplicate_groups(conn)
+        groups, conflicted = [], []
+        for ids in candidates:
+            reason = self._duplicate_conflict(conn, ids)
+            if reason:
+                conflicted.append((ids, reason))
+            else:
+                groups.append(ids)
+        fingerprint = _Fingerprint("duplicate_prompts")
+        fingerprint.add({"groups": groups, "paired": paired, "conflicted": conflicted})
+        _add_prompt_state(fingerprint, conn, [i for group in groups for i in group])
+        return groups, paired, conflicted, fingerprint.hexdigest()
+
+    def _plan_duplicate_prompts(self, conn):
+        groups, paired, conflicted, fingerprint = self._duplicates_fingerprint(conn)
+        removed = sum(len(group) - 1 for group in groups)
+        held = [ids for ids, _reason in conflicted]
+        texts = _texts(conn, [group[0] for group in (groups + paired + held)])
+        details = (
+            [
+                {
+                    "label": _excerpt(texts.get(group[0])),
+                    "detail": f"Keeps the oldest of {len(group)} copies",
+                }
+                for group in groups
+            ]
+            + [
+                {
+                    "label": _excerpt(texts.get(group[0])),
+                    "detail": "Kept: belongs to a saved pair",
+                }
+                for group in paired
+            ]
+            + [
+                {
+                    "label": _excerpt(texts.get(ids[0])),
+                    "detail": CONFLICT_DETAILS[reason],
+                }
+                for ids, reason in conflicted
+            ]
+        )
+        if groups:
+            label = (
+                f"{_plural(removed, 'duplicate prompt')} in "
+                f"{_plural(len(groups), 'group')}"
+            )
+        else:
+            label = "No duplicate prompt text"
+        reasons = []
+        if len(paired) == 1:
+            reasons.append("1 group belongs to a saved pair")
+        elif paired:
+            reasons.append(f"{len(paired):,} groups belong to saved pairs")
+        for reason, (one, many) in CONFLICT_REASONS.items():
+            count = sum(1 for _ids, why in conflicted if why == reason)
+            if count:
+                reasons.append(f"{count:,} {one if count == 1 else many}")
+        return MaintenanceDraft(
+            operation="duplicate_prompts",
+            fingerprint=fingerprint,
+            targets=groups,
+            scope={"count": removed, "label": label},
+            protected={
+                "count": sum(len(group) for group in paired + held),
+                "reasons": reasons,
+            },
+            media={"deletes_files": False, "label": NO_FILES_DELETED},
+            details=details[:MAX_DETAILS],
+        )
+
+    def _unused_fingerprint(self, conn):
+        ids = _orphaned_ids(conn)
+        fingerprint = _Fingerprint("unused_components")
+        fingerprint.add(ids)
+        _add_prompt_state(fingerprint, conn, ids)
+        return ids, fingerprint.hexdigest()
+
+    def _plan_unused_components(self, conn):
+        ids, fingerprint = self._unused_fingerprint(conn)
+        *by_reason, kept = conn.execute(_UNLINKED_PROTECTIONS_SQL).fetchone()
+        reasons = [
+            f"{count:,} {many if count != 1 else one}"
+            for count, (_predicate, many, one) in zip(by_reason, PROTECTIONS)
+            if count
+        ]
+        texts = _texts(conn, ids)
+        count = len(ids)
+        return MaintenanceDraft(
+            operation="unused_components",
+            fingerprint=fingerprint,
+            targets=ids,
+            scope={
+                "count": count,
+                "label": (
+                    _plural(count, "unused prompt component")
+                    if count
+                    else "No unused prompt components"
+                ),
+            },
+            protected={"count": kept, "reasons": reasons},
+            media={"deletes_files": False, "label": NO_FILES_DELETED},
+            details=[
+                {"label": _excerpt(texts.get(prompt_id)), "detail": UNUSED_DETAIL}
+                for prompt_id in ids[:MAX_DETAILS]
+            ],
+        )
+
+    @staticmethod
+    def _links_fingerprint(candidates):
+        fingerprint = _Fingerprint("missing_links")
+        for row, owner in candidates:
+            fingerprint.add([row[0], row[1], row[3], owner])
+        return fingerprint.hexdigest()
+
+    def _plan_missing_links(self, rows, roots):
+        readable = readable_roots(roots)
+        candidates, unavailable, outside = [], 0, 0
+        for row, owner, state in link_states(rows, roots, readable):
+            if state == "missing":
+                candidates.append((row, owner))
+            elif state == "unavailable":
+                unavailable += 1
+            elif state == "outside":
+                outside += 1
+        reasons = []
+        if unavailable:
+            reasons.append(
+                _plural(
+                    unavailable,
+                    "link under an unavailable source",
+                    "links under unavailable sources",
+                )
+            )
+        if outside:
+            reasons.append(
+                _plural(
+                    outside,
+                    "link outside the image sources",
+                    "links outside the image sources",
+                )
+            )
+        count = len(candidates)
+        return MaintenanceDraft(
+            operation="missing_links",
+            fingerprint=self._links_fingerprint(candidates),
+            targets=[row[0] for row, _owner in candidates],
+            scope={
+                "count": count,
+                "label": (
+                    _plural(count, "link to a missing file", "links to missing files")
+                    if count
+                    else "No links to missing files"
+                ),
+            },
+            protected={"count": unavailable + outside, "reasons": reasons},
+            media={"deletes_files": False, "label": LINKS_ONLY},
+            details=[
+                {
+                    "label": _relative(row[1], roots[owner]),
+                    "detail": (
+                        f"{_source_name(roots[owner], owner)} · the file is missing"
+                    ),
+                }
+                for row, owner in candidates[:MAX_DETAILS]
+            ],
+        )
+
+    def _plan_optimize(self):
+        def read(conn):
+            return [
+                conn.execute(f"PRAGMA {name}").fetchone()[0]
+                for name in ("page_size", "page_count", "freelist_count")
+            ]
+
+        page_size, page_count, free_pages = _read_snapshot(self.model.db_path, read)
+        db_path = self.model.db_path
+        wal = db_path + "-wal"
+        size = page_size * page_count
+        wal_size = os.path.getsize(wal) if os.path.exists(wal) else 0
+        free = shutil.disk_usage(os.path.dirname(os.path.abspath(db_path))).free
+        # VACUUM writes the rebuilt database through the WAL beside it.
+        return MaintenanceDraft(
+            operation="optimize",
+            fingerprint=_Fingerprint("optimize").hexdigest(),
+            targets=[],
+            scope={"count": 1, "label": "The whole database"},
+            protected={"count": 0, "reasons": []},
+            media={"deletes_files": False, "label": NO_FILES_DELETED},
+            details=[
+                {
+                    "label": "Database file",
+                    "detail": f"{_megabytes(size)}, plus {_megabytes(wal_size)} "
+                    "not yet checkpointed",
+                },
+                {
+                    "label": "Free pages",
+                    "detail": f"About {_megabytes(free_pages * page_size)} "
+                    "can be reclaimed",
+                },
+                {
+                    "label": "Lock",
+                    "detail": OPTIMIZE_LOCK_NOTE,
+                },
+                {
+                    "label": "Free space needed",
+                    "detail": f"About {_megabytes(size)} beside the database and "
+                    f"{_megabytes(size)} in SQLite's temporary folder; "
+                    f"{_megabytes(free)} available beside the database",
+                },
+            ],
+        )
+
+    # ── apply ─────────────────────────────────────────────────────────
+
+    def apply_maintenance(self, operation, fingerprint, targets, roots=()) -> dict:
+        """Apply a reviewed plan exactly, or raise PlanChanged (blocking).
+
+        Inside ``BEGIN IMMEDIATE`` the fingerprint is computed again on the
+        write connection; when it differs nothing is written. ``targets`` and
+        ``roots`` are the plan's. Returns ``{changed, rows}``: how many
+        records changed and up to ``MAX_RESULT_ROWS`` descriptions of them
+        (``{prompt_id[, kept_id]}`` or ``{image_path}``).
+        """
+        if operation not in MAINTENANCE_OPERATIONS or operation == "workflow_storage":
+            raise ValueError(f"Unknown maintenance operation {operation!r}")
+        if operation == "optimize":
+            # VACUUM cannot run in a transaction; the plan binds the database.
+            if fingerprint != _Fingerprint("optimize").hexdigest():
+                raise PlanChanged("optimize")
+            self.model.vacuum_database()
+            return {"changed": 1, "rows": []}
+        appliers = {
+            "duplicate_prompts": self._apply_duplicate_prompts,
+            "unused_components": self._apply_unused_components,
+            "missing_links": self._apply_missing_links,
+        }
+        with self._pair_transaction() as conn:
+            return appliers[operation](conn, fingerprint, targets, roots)
+
+    def _apply_duplicate_prompts(self, conn, fingerprint, targets, _roots):
+        groups, _paired, _conflicted, current = self._duplicates_fingerprint(conn)
+        if current != fingerprint or groups != targets:
+            raise PlanChanged("duplicate_prompts")
+        changed, rows = 0, []
+        for sorted_ids in groups:
+            removed, _moved = self._merge_duplicate_group(conn, sorted_ids, strict=True)
+            changed += removed
+            rows.extend(
+                {"prompt_id": duplicate_id, "kept_id": sorted_ids[0]}
+                for duplicate_id in sorted_ids[1:]
+            )
+        return {"changed": changed, "rows": rows[:MAX_RESULT_ROWS]}
+
+    def _apply_unused_components(self, conn, fingerprint, targets, _roots):
+        ids, current = self._unused_fingerprint(conn)
+        if current != fingerprint or ids != targets:
+            raise PlanChanged("unused_components")
+        texts = _texts(conn, ids, limit=MAX_RESULT_ROWS)
+        changed = _delete_ids(conn, "prompts", ids)
+        rows = [
+            {"prompt_id": prompt_id, "text": _excerpt(texts.get(prompt_id))}
+            for prompt_id in ids[:MAX_RESULT_ROWS]
+        ]
+        return {"changed": changed, "rows": rows}
+
+    def _apply_missing_links(self, conn, fingerprint, targets, roots):
+        rows = []
+        for batch in _batches(sorted(targets)):
+            rows.extend(
+                tuple(row)
+                for row in conn.execute(
+                    "SELECT id, image_path, filename, file_path FROM generated_images "
+                    f"WHERE id IN ({_marks(batch)}) ORDER BY id",  # nosec B608
+                    batch,
+                )
+            )
+        readable = readable_roots(roots)
+        candidates = []
+        for row, owner, state in link_states(rows, roots, readable):
+            if state != "missing":  # back, or its source went away
+                raise PlanChanged("missing_links")
+            candidates.append((row, owner))
+        if self._links_fingerprint(candidates) != fingerprint:
+            raise PlanChanged("missing_links")
+        changed = _delete_ids(conn, "generated_images", [row[0] for row in rows])
+        return {
+            "changed": changed,
+            "rows": [{"image_path": row[1]} for row in rows[:MAX_RESULT_ROWS]],
+        }

database/managed_connection.py

@@ -0,0 +1,353 @@
+"""A SQLite connection that is never closed under the thread using it.
+
+``PromptModel`` gives each thread its own connection, but Move library and
+``close_all`` must retire every thread's connection from one thread. Closing
+a ``sqlite3.Connection`` while its owner is inside a statement (Python runs
+``sqlite3_step`` without the GIL) frees memory the owner then reads: on
+Windows the process dies with "access violation", elsewhere the owner sees
+garbage such as a ``MemoryError``.
+
+``ManagedConnection`` counts the calls its owner is inside. ``retire()`` is
+safe from any thread and never raises: an idle connection is closed at once;
+one in use is closed by its owner as the call ends. A retired connection
+refuses new work (cursors, statements, commits) with the error a closed
+connection raises.
+
+What retiring a connection in use does and does not do:
+
+- It calls ``sqlite3_interrupt`` (thread-safe). That ends a running
+  statement at its next step, but it does not wake a statement that is
+  waiting for another connection's lock: that one waits until the lock is
+  released (or the busy timeout ends it), and only then is it interrupted.
+  An interrupt that arrives between two statements, for example between the
+  implicit ``BEGIN`` and the DML or between ``executemany`` rows, is lost.
+- Writes are kept out of a file the process has switched away from because
+  a retired connection never commits: legacy-mode DML runs inside a
+  transaction that only ``commit()`` (refused) or ``close()`` (rolls back)
+  ends. A statement that commits by itself is not covered: DDL or a PRAGMA
+  write outside a transaction, or an explicit ``COMMIT`` passed to
+  ``execute``/``executescript`` (``executescript`` also commits first, in C,
+  on Python before 3.12). Managed connections must never be used that way
+  for library writes.
+"""
+
+from __future__ import annotations
+
+import contextlib
+import logging
+import threading
+import time
+import weakref
+from typing import Iterable, Iterator, List, Optional
+
+import sqlite3
+
+CLOSED_MESSAGE = "Cannot operate on a closed database."
+# Rows fetched per step when a cursor is iterated: the connection is marked
+# in use once per chunk instead of once per row (iteration costs about what
+# a plain cursor costs). A loop that stops early leaves up to this many rows
+# fetched; ``fetchone``/``fetchmany`` after a partial loop skip them.
+ITER_CHUNK = 128
+
+
+class ManagedCursor(sqlite3.Cursor):
+    """A cursor whose calls into SQLite count as its connection being in use."""
+
+    def execute(self, sql, parameters=(), /):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().execute(sql, parameters)
+        finally:
+            conn._leave()
+
+    def executemany(self, sql, seq_of_parameters, /):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().executemany(sql, seq_of_parameters)
+        finally:
+            conn._leave()
+
+    def executescript(self, sql_script, /):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().executescript(sql_script)
+        finally:
+            conn._leave()
+
+    def fetchone(self):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().fetchone()
+        finally:
+            conn._leave()
+
+    def fetchmany(self, *args, **kwargs):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().fetchmany(*args, **kwargs)
+        finally:
+            conn._leave()
+
+    def fetchall(self):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().fetchall()
+        finally:
+            conn._leave()
+
+    def __iter__(self) -> Iterator:
+        return self._chunks()
+
+    def _chunks(self) -> Iterator:
+        while True:
+            rows = self.fetchmany(ITER_CHUNK)
+            if not rows:
+                return
+            yield from rows
+
+    def __next__(self):
+        conn = self.connection
+        conn._enter()
+        try:
+            return super().__next__()
+        finally:
+            conn._leave()
+
+    def close(self):
+        conn = self.connection
+        try:
+            conn._enter()
+        except sqlite3.ProgrammingError:
+            return super().close()  # the connection is closed or closing
+        try:
+            return super().close()
+        finally:
+            conn._leave()
+
+
+class ManagedConnection(sqlite3.Connection):
+    """A connection only its owner closes while a call into it is running."""
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        # Guards _depth, _retired, _closing and _cursors. Re-entrant so that
+        # code running in the closing thread while it holds the lock (a
+        # cursor() during the cursor-set copy) is refused instead of hanging.
+        self._use_lock = threading.RLock()
+        self._depth = 0
+        self._retired = False
+        self._closing = False
+        self._released = threading.Event()
+        # Python before 3.12 keeps the file open after close() while a cursor
+        # still holds a statement; these are closed first (see _close_now).
+        self._cursors: "weakref.WeakSet[sqlite3.Cursor]" = weakref.WeakSet()
+        self.close_logger: Optional[logging.Logger] = None
+
+    # ── in use ───────────────────────────────────────────────────────
+
+    def _enter(self) -> None:
+        with self._use_lock:
+            if self._retired:
+                raise sqlite3.ProgrammingError(CLOSED_MESSAGE)
+            self._depth += 1
+
+    def _leave(self) -> None:
+        with self._use_lock:
+            self._depth -= 1
+            close_now = self._retired and self._depth == 0
+        if close_now:
+            self._close_now()
+
+    @contextlib.contextmanager
+    def using(self):
+        """Mark the connection in use for a call this class cannot wrap.
+
+        For example as the *target* of another connection's ``backup()``.
+        """
+        self._enter()
+        try:
+            yield self
+        finally:
+            self._leave()
+
+    def cursor(self, factory=None):
+        with self._use_lock:
+            if self._retired:
+                raise sqlite3.ProgrammingError(CLOSED_MESSAGE)
+            cursor = super().cursor(factory or ManagedCursor)
+            self._cursors.add(cursor)
+        return cursor
+
+    def execute(self, sql, parameters=(), /):
+        return self.cursor().execute(sql, parameters)
+
+    def executemany(self, sql, seq_of_parameters, /):
+        return self.cursor().executemany(sql, seq_of_parameters)
+
+    def executescript(self, sql_script, /):
+        return self.cursor().executescript(sql_script)
+
+    def commit(self):
+        self._enter()
+        try:
+            return super().commit()
+        finally:
+            self._leave()
+
+    def rollback(self):
+        self._enter()  # a closed connection raises, as sqlite3's does
+        try:
+            return super().rollback()
+        finally:
+            self._leave()
+
+    def backup(self, target, **kwargs):
+        self._enter()
+        try:
+            return super().backup(target, **kwargs)
+        finally:
+            self._leave()
+
+    def create_function(self, *args, **kwargs):
+        self._enter()
+        try:
+            return super().create_function(*args, **kwargs)
+        finally:
+            self._leave()
+
+    @property
+    def in_transaction(self) -> bool:
+        self._enter()
+        try:
+            return super().in_transaction
+        finally:
+            self._leave()
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        # The C implementation calls its own commit/rollback, not these.
+        if exc_type is None:
+            try:
+                self.commit()
+            except BaseException:
+                with contextlib.suppress(sqlite3.Error):
+                    self.rollback()
+                raise
+        else:
+            self.rollback()
+        return False
+
+    # ── closing ──────────────────────────────────────────────────────
+
+    def close(self) -> None:
+        """Close now if idle, else as soon as the running call ends.
+
+        Unlike ``sqlite3.Connection.close``, a call from another thread while
+        the owner is inside a call returns before the file is closed. A caller
+        that must know the file is closed (before renaming or deleting it)
+        waits for ``wait_closed`` (``PromptModel.close_detached`` does).
+        """
+        self.retire()
+
+    def retire(self) -> bool:
+        """Refuse new work and close; True when the connection is closed already.
+
+        Safe from any thread, and never raises: failures are logged.
+        """
+        try:
+            with self._use_lock:
+                if not self._retired:
+                    self._retired = True
+                    if self._depth:
+                        # Ends the running statement (see the module docstring
+                        # for what it does not end). The owner cannot start
+                        # closing the connection while this lock is held.
+                        self.interrupt()
+                        return False
+                elif self._depth:
+                    return False
+                else:
+                    return self._released.wait(0)
+        except Exception:
+            self._log_exception("Error retiring database connection")
+            return self._released.wait(0)
+        self._close_now()
+        return True
+
+    def wait_closed(self, timeout: float) -> bool:
+        return self._released.wait(timeout)
+
+    def _close_now(self) -> None:
+        # Only ever called with no call into the connection running, and a
+        # retired connection refuses new calls and cursors: its cursors are
+        # idle, and the set is copied under the lock that guards it.
+        with self._use_lock:
+            if self._closing:
+                return  # another thread closes it and sets _released
+            self._closing = True
+            cursors = list(self._cursors)
+        try:
+            for cursor in cursors:
+                with contextlib.suppress(sqlite3.Error):
+                    sqlite3.Cursor.close(cursor)
+            super().close()
+        except Exception:
+            self._log_exception("Error closing database connection")
+        finally:
+            self._released.set()
+
+    def _log_exception(self, message: str) -> None:
+        if self.close_logger is not None:
+            self.close_logger.exception(message)
+
+
+def retire_connections(
+    connections: Iterable[sqlite3.Connection], logger=None
+) -> List[ManagedConnection]:
+    """Retire every connection; return those whose owner has yet to close them.
+
+    Safe while holding a lock the owners wait on: nothing here waits for them.
+    One connection failing to retire is logged and never skips the rest.
+    """
+    pending = []
+    for conn in connections:
+        try:
+            if isinstance(conn, ManagedConnection):
+                if logger is not None:
+                    conn.close_logger = logger
+                if not conn.retire():
+                    pending.append(conn)
+            else:
+                conn.close()
+        except Exception:
+            if logger is not None:
+                logger.exception("Error closing database connection")
+    return pending
+
+
+def await_closed(
+    connections: Iterable[ManagedConnection], timeout: float, logger=None
+) -> bool:
+    """Wait until the owners closed ``connections`` (retired); False on timeout.
+
+    A connection still open after ``timeout`` (its owner is blocked, for
+    example on another process's lock after a lost interrupt) is logged and
+    left to close when its call ends; a Move library then reports the old
+    files as still in use instead of renaming them.
+    """
+    deadline = time.monotonic() + timeout
+    still_open = 0
+    for conn in connections:
+        if not conn.wait_closed(max(0.0, deadline - time.monotonic())):
+            still_open += 1
+    if still_open and logger is not None:
+        logger.warning(
+            f"{still_open} database connection(s) still in use after "
+            f"{timeout:g}s; they close when their statement ends"
+        )
+    return not still_open

database/models.py

@@ -4,14 +4,42 @@ Database schema and models for KikoTextEncode prompt storage.
 
 import contextlib
 import datetime
-import shutil
+import json
+import math
 import sqlite3
 import os
 import threading
 import weakref
 from contextlib import closing
 from pathlib import Path
-from typing import Set, Tuple
+from typing import Optional, Set, Tuple
+
+from .managed_connection import (
+    ManagedConnection,
+    await_closed,
+    retire_connections,
+)
+from . import restore_stage
+from .library_schema import library_schema_error, migrate_library_state
+from .image_favorites import image_favorites_schema_error, migrate_image_favorites
+from .annotation_schema import annotation_schema_error, migrate_annotation_state
+from .review_schema import (
+    migrate_review_state,
+    review_schema_error,
+    review_without_annotations_error,
+)
+from .saved_views import migrate_saved_views, saved_views_schema_error
+from .workflow_documents import (
+    DOCUMENT_ID_COLUMNS,
+    migrate_workflow_documents,
+    workflow_documents_schema_error,
+)
+
+
+from .execution_history_schema import (
+    history_schema_error,
+    migrate_execution_history,
+)
 
 # Import logging system
 try:
@@ -29,6 +57,54 @@ except ImportError:
 ISO_TIMESTAMP_FORMAT = "%Y-%m-%dT%H:%M:%f+00:00"
 
 
+MAX_PAIR_PROVENANCE_LENGTH = 4_000_000
+MAX_PAIR_PROVENANCE_DEPTH = 64
+
+
+def _refuse_rebuild_with_documents(columns) -> None:
+    """The legacy generated_images rebuilds copy a fixed column list.
+
+    A file that needs one (``prompt_id`` TEXT, no ``file_path``) but already
+    has the workflow document ids would lose them, so it is refused.
+    """
+    if any(column in columns for column in DOCUMENT_ID_COLUMNS):
+        raise sqlite3.DatabaseError(
+            "This database needs an old image table rebuild but already has "
+            "workflow document references; the rebuild would drop them"
+        )
+
+
+def pair_provenance_error(encoded: object) -> Optional[str]:
+    """Validate the same bounded JSON object for live writes and restores."""
+    if not isinstance(encoded, str):
+        return "Pair provenance must be JSON text"
+    if len(encoded) > MAX_PAIR_PROVENANCE_LENGTH:
+        return "Pair provenance exceeds the size limit"
+    try:
+        value = json.loads(encoded)
+    except (ValueError, TypeError, RecursionError):
+        return "Pair provenance is not readable JSON"
+    if not isinstance(value, dict):
+        return "Pair provenance must be a JSON object"
+    pending = [(iter((value,)), 0)]
+    while pending:
+        children, depth = pending[-1]
+        try:
+            item = next(children)
+        except StopIteration:
+            pending.pop()
+            continue
+        if depth > MAX_PAIR_PROVENANCE_DEPTH:
+            return "Pair provenance exceeds the nesting limit"
+        if isinstance(item, float) and not math.isfinite(item):
+            return "Pair provenance contains nonfinite numbers"
+        if isinstance(item, dict):
+            pending.append((iter(item.values()), depth + 1))
+        elif isinstance(item, list):
+            pending.append((iter(item), depth + 1))
+    return None
+
+
 def utc_now_iso() -> str:
     """Current UTC time as ISO text with microseconds.
 
@@ -57,6 +133,24 @@ SQLITE_HEADER = b"SQLite format 3\x00"
 REQUIRED_PROMPT_COLUMNS = ("id", "text", "created_at")
 
 
+def configure_connection(conn: sqlite3.Connection) -> None:
+    """Apply the locking and durability settings every managed connection shares.
+
+    WAL lets readers run beside the single writer. In WAL mode
+    ``synchronous=NORMAL`` keeps the database consistent after an application
+    or ComfyUI crash and flushes only at checkpoints, so a commit no longer
+    holds SQLite's write lock across a disk flush. Under ``FULL`` a slow flush
+    (common on Windows) let other writers' polling busy handlers starve past
+    BUSY_TIMEOUT_SECONDS and fail with "database is locked" (#242). Without
+    WAL (for example an in-memory database) SQLite's FULL default is kept.
+    """
+    journal_mode = conn.execute("PRAGMA journal_mode = WAL").fetchone()[0]
+    conn.execute("PRAGMA foreign_keys = ON")
+    conn.execute(f"PRAGMA busy_timeout = {BUSY_TIMEOUT_SECONDS * 1000}")
+    if str(journal_mode).lower() == "wal":
+        conn.execute("PRAGMA synchronous = NORMAL")
+
+
 class PromptModel:
     """Database model for prompt storage and schema management."""
 
@@ -81,7 +175,7 @@ class PromptModel:
         self.logger.debug(f"Initializing database model with path: {db_path}")
         self.db_path = db_path
         # One connection per thread (see get_connection); every open connection
-        # is also tracked here so restore_from_file can close them all.
+        # is also tracked here so close_all can clean up across threads.
         self._local = threading.local()
         self._connections: Set[sqlite3.Connection] = set()
         self._conn_lock = threading.Lock()
@@ -109,9 +203,7 @@ class PromptModel:
             try:
                 conn = sqlite3.connect(self.db_path, timeout=BUSY_TIMEOUT_SECONDS)
                 try:
-                    conn.execute("PRAGMA journal_mode = WAL")
-                    conn.execute(f"PRAGMA busy_timeout = {BUSY_TIMEOUT_SECONDS * 1000}")
-                    conn.execute("PRAGMA foreign_keys = ON")
+                    configure_connection(conn)
                     self._create_tables(conn)
                     self._create_indexes(conn)
                     conn.commit()
@@ -235,6 +327,241 @@ class PromptModel:
 
         # Image uniqueness by full path instead of basename (3.2.4)
         self._migrate_image_uniqueness_by_path(conn)
+        self._migrate_prompt_pairs(conn)
+        migrate_execution_history(conn)
+        migrate_library_state(conn)
+        migrate_image_favorites(conn)
+        migrate_saved_views(conn)
+        # A review group without its annotation group is refused before the
+        # annotation install could make it look valid (catalog only; the rows
+        # are validated once, in migrate_review_state).
+        review_error = review_without_annotations_error(conn)
+        if review_error:
+            raise sqlite3.DatabaseError(review_error)
+        migrate_annotation_state(conn)
+        migrate_review_state(conn)
+        # Last: the older generated_images rebuilds run only before file_path.
+        migrate_workflow_documents(conn)
+
+    def _migrate_prompt_pairs(self, conn: sqlite3.Connection) -> None:
+        conn.execute("SAVEPOINT prompt_pair_schema")
+        try:
+            conn.execute("""
+                CREATE TABLE IF NOT EXISTS prompt_pairs (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    positive_prompt_id INTEGER NOT NULL,
+                    negative_prompt_id INTEGER,
+                    title TEXT,
+                    category TEXT,
+                    rating INTEGER CHECK(rating BETWEEN 1 AND 5),
+                    notes TEXT,
+                    created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+                    updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+                    last_used_at TEXT,
+                    run_count INTEGER NOT NULL DEFAULT 0,
+                    archived INTEGER NOT NULL DEFAULT 0 CHECK(archived IN (0, 1)),
+                    FOREIGN KEY(positive_prompt_id)
+                        REFERENCES prompts(id) ON DELETE RESTRICT,
+                    FOREIGN KEY(negative_prompt_id)
+                        REFERENCES prompts(id) ON DELETE RESTRICT,
+                    UNIQUE(positive_prompt_id, negative_prompt_id)
+                )
+            """)
+            conn.execute("""
+                CREATE TABLE IF NOT EXISTS prompt_pair_tags (
+                    pair_id INTEGER NOT NULL,
+                    tag_id INTEGER NOT NULL,
+                    PRIMARY KEY(pair_id, tag_id),
+                    FOREIGN KEY(pair_id) REFERENCES prompt_pairs(id) ON DELETE CASCADE,
+                    FOREIGN KEY(tag_id) REFERENCES tags(id) ON DELETE CASCADE
+                )
+            """)
+            conn.execute("""
+                CREATE TABLE IF NOT EXISTS image_prompt_pairs (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    image_id INTEGER NOT NULL,
+                    pair_id INTEGER NOT NULL,
+                    provenance TEXT NOT NULL,
+                    created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+                    FOREIGN KEY(image_id)
+                        REFERENCES generated_images(id) ON DELETE CASCADE,
+                    FOREIGN KEY(pair_id) REFERENCES prompt_pairs(id) ON DELETE RESTRICT,
+                    UNIQUE(pair_id, image_id, provenance)
+                )
+            """)
+            self._create_pair_indexes(conn)
+            error = self._pair_schema_error(conn)
+            if error:
+                raise sqlite3.DatabaseError(error)
+            conn.execute("RELEASE prompt_pair_schema")
+        except BaseException:
+            conn.execute("ROLLBACK TO prompt_pair_schema")
+            conn.execute("RELEASE prompt_pair_schema")
+            raise
+
+    @staticmethod
+    def _create_pair_indexes(conn: sqlite3.Connection) -> None:
+        conn.execute(
+            "CREATE UNIQUE INDEX IF NOT EXISTS idx_prompt_pairs_positive_only "
+            "ON prompt_pairs(positive_prompt_id) WHERE negative_prompt_id IS NULL"
+        )
+        for name, target in (
+            ("idx_prompt_pairs_negative", "prompt_pairs(negative_prompt_id)"),
+            ("idx_prompt_pair_tags_tag", "prompt_pair_tags(tag_id)"),
+            ("idx_image_prompt_pairs_image", "image_prompt_pairs(image_id)"),
+        ):
+            conn.execute(f"CREATE INDEX IF NOT EXISTS {name} ON {target}")
+
+    @staticmethod
+    def _pair_schema_error(conn: sqlite3.Connection) -> Optional[str]:
+        required = {
+            "prompt_pairs": {
+                "id",
+                "positive_prompt_id",
+                "negative_prompt_id",
+                "title",
+                "category",
+                "rating",
+                "notes",
+                "created_at",
+                "updated_at",
+                "last_used_at",
+                "run_count",
+                "archived",
+            },
+            "prompt_pair_tags": {"pair_id", "tag_id"},
+            "image_prompt_pairs": {
+                "id",
+                "image_id",
+                "pair_id",
+                "provenance",
+                "created_at",
+            },
+        }
+        present = {
+            row[0]
+            for row in conn.execute("SELECT name FROM sqlite_master WHERE type='table'")
+        }
+        if not present.intersection(required):
+            return None
+        if not set(required).issubset(present):
+            return "Incomplete prompt pair schema"
+        foreign_keys = {
+            "prompt_pairs": {
+                ("positive_prompt_id", "prompts", "id", "RESTRICT"),
+                ("negative_prompt_id", "prompts", "id", "RESTRICT"),
+            },
+            "prompt_pair_tags": {
+                ("pair_id", "prompt_pairs", "id", "CASCADE"),
+                ("tag_id", "tags", "id", "CASCADE"),
+            },
+            "image_prompt_pairs": {
+                ("pair_id", "prompt_pairs", "id", "RESTRICT"),
+                ("image_id", "generated_images", "id", "CASCADE"),
+            },
+        }
+        unique = {
+            "prompt_pairs": ("positive_prompt_id", "negative_prompt_id"),
+            "prompt_pair_tags": ("pair_id", "tag_id"),
+            "image_prompt_pairs": ("pair_id", "image_id", "provenance"),
+        }
+        non_null = {
+            "prompt_pairs": {
+                "positive_prompt_id",
+                "created_at",
+                "updated_at",
+                "run_count",
+                "archived",
+            },
+            "prompt_pair_tags": {"pair_id", "tag_id"},
+            "image_prompt_pairs": {"pair_id", "image_id", "provenance", "created_at"},
+        }
+        integer_columns = {
+            "id",
+            "positive_prompt_id",
+            "negative_prompt_id",
+            "pair_id",
+            "tag_id",
+            "image_id",
+            "rating",
+            "run_count",
+            "archived",
+        }
+        for table, columns in required.items():
+            info = {row[1]: row for row in conn.execute(f"PRAGMA table_info({table})")}
+            if not columns.issubset(info):
+                return f"{table} is missing required pair columns"
+            if any(not info[name][3] for name in non_null[table]):
+                return f"{table} has nullable required pair columns"
+            if any(
+                info[name][2].upper()
+                != ("INTEGER" if name in integer_columns else "TEXT")
+                for name in columns
+            ):
+                return f"{table} has invalid pair column types"
+            primary = tuple(
+                name
+                for name in sorted(info, key=lambda name: info[name][5])
+                if info[name][5]
+            )
+            if primary != (
+                ("pair_id", "tag_id") if table == "prompt_pair_tags" else ("id",)
+            ):
+                return f"{table} has an invalid pair primary key"
+            actual_fks = {
+                (row[3], row[2], row[4], row[6])
+                for row in conn.execute(f"PRAGMA foreign_key_list({table})")
+            }
+            if actual_fks != foreign_keys[table]:
+                return f"{table} has invalid pair foreign key definitions"
+            if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
+                return f"{table} contains foreign key violations"
+            indexes = list(conn.execute(f"PRAGMA index_list({table})"))
+            index_columns = {}
+            for row in indexes:
+                name = row[1].replace('"', '""')
+                index_columns[row[1]] = tuple(
+                    item[2] for item in conn.execute(f'PRAGMA index_info("{name}")')
+                )
+            if not any(
+                row[2] and not row[4] and index_columns[row[1]] == unique[table]
+                for row in indexes
+            ):
+                return f"{table} is missing pair uniqueness constraints"
+            if table == "prompt_pairs":
+                null_unique = False
+                for row in indexes:
+                    if (
+                        row[2]
+                        and row[4]
+                        and index_columns[row[1]] == ("positive_prompt_id",)
+                    ):
+                        sql = conn.execute(
+                            "SELECT sql FROM sqlite_master WHERE type='index' AND name"
+                            "=?",
+                            (row[1],),
+                        ).fetchone()[0]
+                        if " ".join(sql.lower().split()).endswith(
+                            "where negative_prompt_id is null"
+                        ):
+                            null_unique = True
+                if not null_unique:
+                    return "prompt_pairs is missing empty-negative pair uniqueness"
+        if conn.execute(
+            "SELECT 1 FROM image_prompt_pairs ip "
+            "JOIN prompt_pairs pp ON pp.id=ip.pair_id "
+            "JOIN generated_images gi ON gi.id=ip.image_id "
+            "WHERE pp.positive_prompt_id IS NOT gi.prompt_id LIMIT 1"
+        ).fetchone():
+            return "Image pair association has a different positive prompt owner"
+        # fetchone, not iteration: a managed connection's cursor prefetches
+        # ITER_CHUNK rows when iterated, and each provenance may be ~4 MB.
+        cursor = conn.execute("SELECT provenance FROM image_prompt_pairs")
+        while (row := cursor.fetchone()) is not None:
+            error = pair_provenance_error(row[0])
+            if error:
+                return error
+        return None
 
     def _create_indexes(self, conn: sqlite3.Connection) -> None:
         """
@@ -274,10 +601,11 @@ class PromptModel:
 
         Each thread gets its own ``sqlite3.Connection`` (kept in
         ``threading.local``), created on first use and reused by that thread
-        thereafter. Every connection is configured with WAL journaling,
-        ``row_factory = sqlite3.Row``, ``PRAGMA foreign_keys = ON`` and
-        ``PRAGMA busy_timeout`` of BUSY_TIMEOUT_SECONDS so concurrent writers
-        wait instead of failing. Callers use it as a context manager
+        thereafter. Every connection gets ``row_factory = sqlite3.Row`` and the
+        shared settings of ``configure_connection``: WAL journaling with
+        ``synchronous=NORMAL``, ``PRAGMA foreign_keys = ON`` and a
+        ``busy_timeout`` of BUSY_TIMEOUT_SECONDS so concurrent writers wait
+        instead of failing. Callers use it as a context manager
         (``with conn:``), which commits or rolls back that thread's transaction
         and leaves the connection open. No lock is held while a connection is in use.
 
@@ -288,21 +616,42 @@ class PromptModel:
         if conn is not None and self._local.generation == self._generation:
             return conn
 
-        # check_same_thread=False only so restore_from_file can close other
-        # threads' connections; each connection is still used by one thread.
-        conn = sqlite3.connect(
-            self.db_path, check_same_thread=False, timeout=BUSY_TIMEOUT_SECONDS
-        )
-        conn.row_factory = sqlite3.Row
-        conn.execute("PRAGMA journal_mode = WAL")
-        conn.execute("PRAGMA foreign_keys = ON")
-        conn.execute(f"PRAGMA busy_timeout = {BUSY_TIMEOUT_SECONDS * 1000}")
+        while True:
+            # The path and generation are read together, and the new
+            # connection is kept only if no Move library repointed the model
+            # meanwhile: otherwise it would write to the old file.
+            with self._conn_lock:
+                path, generation = self.db_path, self._generation
+            # check_same_thread=False only so close_all and Move library can
+            # retire other threads' connections; each connection is still
+            # used by one thread, and ManagedConnection never closes it under
+            # a running statement (see managed_connection).
+            conn = sqlite3.connect(
+                path,
+                check_same_thread=False,
+                timeout=BUSY_TIMEOUT_SECONDS,
+                factory=ManagedConnection,
+            )
+            try:
+                conn.row_factory = sqlite3.Row
+                configure_connection(conn)
+            except BaseException:
+                conn.close()
+                raise
+            with self._conn_lock:
+                if generation == self._generation:
+                    self._connections.add(conn)
+                    break
+            conn.close()
         self._local.conn = conn
-        self._local.generation = self._generation
-        with self._conn_lock:
-            self._connections.add(conn)
+        self._local.generation = generation
         return conn
 
+    @property
+    def generation(self) -> int:
+        """Bumped whenever every connection is closed or the model repointed."""
+        return self._generation
+
     @classmethod
     def close_all_instances(cls) -> None:
         """Close every connection of every live model in this process."""
@@ -311,22 +660,76 @@ class PromptModel:
         for model in models:
             model.close_all()
 
+    @classmethod
+    def repoint_instances(cls, old_path: str, new_path: str) -> list:
+        """Point every model of ``old_path`` at ``new_path`` (Move library).
+
+        Every thread opens the new file on its next ``get_connection``. The
+        old connections are detached, not closed, and returned: the caller
+        retires them (``retire_detached``) while it still holds the old
+        file's write lock. A write waiting on that lock is then interrupted
+        and can never commit, so it fails (and, through
+        ``PromptDatabase.survive_library_switch``, runs again on the new file).
+        """
+        old_key = os.path.normcase(os.path.realpath(old_path))
+        with cls._instances_lock:
+            models = list(cls._instances)
+        with cls._init_lock:
+            cls._initialized_paths.add(os.path.normcase(os.path.realpath(new_path)))
+        detached = []
+        for model in models:
+            if model._schema_key() == old_key:
+                detached.extend(model._detach_connections(new_path))
+        return detached
+
+    @staticmethod
+    def retire_detached(connections, logger=None) -> list:
+        """Retire detached connections without waiting for their owners.
+
+        Idle ones close now. One whose owner thread is inside a statement is
+        never closed from here (that frees memory the owner is using: an
+        access violation on Windows); it is interrupted, refuses to commit and
+        is closed by its owner as the statement ends. Returns those still
+        open, for ``await_detached`` once no lock the owners wait on is held.
+        """
+        return retire_connections(connections, logger)
+
+    @staticmethod
+    def await_detached(connections, logger=None) -> bool:
+        """Wait until retired connections are closed; False if some are not."""
+        return await_closed(connections, BUSY_TIMEOUT_SECONDS, logger)
+
+    @staticmethod
+    def close_detached(connections, logger=None) -> bool:
+        """Retire connections ``repoint_instances`` detached and await them.
+
+        Do not call it while holding a lock their owners may wait on; retire
+        them there (``retire_detached``) and await them after releasing it.
+        """
+        pending = retire_connections(connections, logger)
+        return await_closed(pending, BUSY_TIMEOUT_SECONDS, logger)
+
     def close_all(self) -> None:
         """Close every thread's connection; they reopen lazily on next use.
 
         Needed before deleting the database file (Windows keeps open files
         locked) and by tests that ran queries through executor threads.
         """
+        self.close_detached(self._detach_connections(), self.logger)
+
+    def _detach_connections(self, new_path: Optional[str] = None) -> list:
+        """Forget every connection (optionally pointing the model elsewhere).
+
+        Returns them for the caller to close; threads reopen on next use.
+        """
         with self._conn_lock:
+            if new_path is not None:
+                self.db_path = new_path
             connections = list(self._connections)
             self._connections.clear()
             self._generation += 1
         self._local.conn = None
-        for conn in connections:
-            try:
-                conn.close()
-            except sqlite3.Error as e:
-                self.logger.warning(f"Error closing database connection: {e}")
+        return connections
 
     def close(self) -> None:
         """Close the calling thread's connection, if it has one.
@@ -340,10 +743,7 @@ class PromptModel:
         self._local.conn = None
         with self._conn_lock:
             self._connections.discard(conn)
-        try:
-            conn.close()
-        except sqlite3.Error as e:
-            self.logger.warning(f"Error closing database connection: {e}")
+        self.close_detached([conn], self.logger)
 
     def _migrate_workflow_name_removal(self, conn: sqlite3.Connection) -> None:
         """
@@ -414,6 +814,7 @@ class PromptModel:
             columns = {column[1]: column[2] for column in cursor.fetchall()}
             if columns.get("prompt_id") != "TEXT":
                 return
+            _refuse_rebuild_with_documents(columns)
 
             self.logger.info("Migrating foreign key types: prompt_id TEXT -> INTEGER")
             with self._table_rebuild(conn):
@@ -579,6 +980,7 @@ class PromptModel:
             }
             if "file_path" in columns:
                 return
+            _refuse_rebuild_with_documents(columns)
 
             self.logger.info("Migrating database: image uniqueness by full path")
             with self._table_rebuild(conn):
@@ -644,6 +1046,10 @@ class PromptModel:
 
         Reclaims unused space and defragments the database file,
         improving query performance and reducing file size.
+
+        Raises:
+            Exception: whatever stopped VACUUM (logged first); the database is
+            then unchanged.
         """
         try:
             with closing(sqlite3.connect(self.db_path)) as conn:
@@ -651,6 +1057,7 @@ class PromptModel:
                 conn.commit()
         except Exception as e:
             self.logger.error(f"Error vacuuming database: {e}")
+            raise
 
     def get_database_info(self) -> dict:
         """
@@ -726,7 +1133,9 @@ class PromptModel:
             return False
 
     @staticmethod
-    def verify_database_file(path: str) -> Tuple[bool, str]:
+    def verify_database_file(
+        path: str, *, require_library_state: bool = False, documents: bool = True
+    ) -> Tuple[bool, str]:
         """
         Check that a file is a healthy prompts database before restoring it.
 
@@ -735,6 +1144,13 @@ class PromptModel:
 
         Args:
             path: File to inspect
+            require_library_state: Also require the tables every current
+                database has (Library state, image favorites, saved views,
+                image annotations and the review queue), as a migrated
+                restore stage must; old backups may lack them.
+            documents: Read every workflow document and refuse a damaged
+                one. A restore checks its upload without it and repairs
+                damaged documents on the stage (``database/restore_stage.py``).
 
         Returns:
             (ok, reason): reason is "ok" when the file is acceptable, otherwise
@@ -772,6 +1188,48 @@ class PromptModel:
                     False,
                     "prompts table is missing required columns: " + ", ".join(missing),
                 )
+            forbidden = conn.execute(
+                "SELECT type FROM sqlite_master WHERE type IN ('trigger', 'view') "
+                "LIMIT 1"
+            ).fetchone()
+            if forbidden:
+                return False, f"Database contains an unsupported {forbidden[0]}"
+            pair_error = PromptModel._pair_schema_error(conn)
+            if pair_error:
+                return False, pair_error
+            history_error = history_schema_error(conn)
+            if history_error:
+                return False, history_error
+            library_error = library_schema_error(
+                conn, allow_missing=not require_library_state
+            )
+            if library_error:
+                return False, library_error
+            favorites_error = image_favorites_schema_error(
+                conn, allow_missing=not require_library_state
+            )
+            if favorites_error:
+                return False, favorites_error
+            views_error = saved_views_schema_error(
+                conn, allow_missing=not require_library_state
+            )
+            if views_error:
+                return False, views_error
+            annotation_error = annotation_schema_error(
+                conn, allow_missing=not require_library_state
+            )
+            if annotation_error:
+                return False, annotation_error
+            review_error = review_schema_error(
+                conn, allow_missing=not require_library_state
+            )
+            if review_error:
+                return False, review_error
+            documents_error = workflow_documents_schema_error(
+                conn, allow_missing=not require_library_state, documents=documents
+            )
+            if documents_error:
+                return False, documents_error
             return True, "ok"
         except sqlite3.Error as e:
             return False, f"Invalid SQLite database: {e}"
@@ -786,41 +1244,59 @@ class PromptModel:
         PromptModel instances that share the file (nodes, image monitor,
         queue hook, API) simply see the new pages on their next statement.
         Nothing is swapped underneath open handles and no -wal/-shm file has
-        to be deleted, which also keeps Windows happy. When SQLite refuses
-        (page sizes differ while the target is in WAL mode) fall back to a
-        file copy with every connection of this instance closed first.
+        to be deleted, which also keeps Windows happy. If SQLite refuses
+        replacement (for example, incompatible page sizes in WAL mode),
+        propagate the error rather than copy files under other open owners.
         """
         source = sqlite3.connect(
             Path(src_path).resolve().as_uri() + "?mode=ro", uri=True
         )
         try:
-            try:
-                source.backup(self.get_connection())
-                return
-            except sqlite3.OperationalError as e:
-                self.logger.warning(
-                    f"Online restore failed ({e}); falling back to a file copy"
-                )
+            target = self.get_connection()
+            # The target is written under backup's step too: mark it in use
+            # so a concurrent close_all retires it instead of closing it.
+            with target.using():
+                source.backup(target)
         finally:
             source.close()
-        self.close_all()
-        shutil.copy2(src_path, self.db_path)
-        for suffix in ("-wal", "-shm"):
-            stale = self.db_path + suffix
-            if os.path.exists(stale):
-                os.remove(stale)
+
+    def stage_restore(self, src_path: str):
+        """Stage ``src_path`` beside this database for a restore (blocking).
+
+        Returns a ``StagedRestore`` (``database/restore_stage.py``): a
+        migrated, repaired and verified copy in the library's folder, its
+        digest and what it holds. Neither the live database nor the source
+        is touched. Raises ValueError when either does not verify.
+        """
+        return restore_stage.stage_restore(PromptModel, self.db_path, src_path)
+
+    def publish_staged(self, staged) -> str:
+        """Replace the live database with ``staged``; return the safety backup.
+
+        See ``restore_stage.publish_staged``: the staged bytes are rechecked,
+        the current database is kept as ``<db_path>.backup_<stamp>`` and the
+        WAL is truncated afterwards.
+        """
+        backup_path = restore_stage.publish_staged(self, staged)
+        self.logger.info("Database restored from a staged copy")
+        from .execution_history import notify_history_restore
+
+        notify_history_restore(self.db_path)
+        return backup_path
 
     def restore_from_file(self, src_path: str) -> str:
         """
         Replace the live database with the file at src_path.
 
-        The current database is first backed up next to itself as
-        ``<db_path>.backup_<YYYYmmdd_HHMMSS>`` (WAL content included), the
-        pages of ``src_path`` are copied into the live database through
-        SQLite's backup API (visible to every open connection) and the schema
-        migrations run on the restored content.
-
-        Callers should run verify_database_file(src_path) first.
+        ``stage_restore`` then ``publish_staged``: a disposable copy of the
+        source is migrated, repaired and verified in the library's folder
+        first. Only then is the current database backed up next to itself as
+        ``<db_path>.backup_<YYYYmmdd_HHMMSS_ffffff>`` (WAL content included),
+        and the staged pages replace the live pages through SQLite's backup
+        API. Staging failures never replace live content or modify the
+        source. An online replacement failure leaves the live database intact
+        and propagates; the completed safety backup remains available. The
+        stage is removed either way.
 
         Args:
             src_path: SQLite file to restore
@@ -830,20 +1306,14 @@ class PromptModel:
             there was no database to back up.
 
         Raises:
-            RuntimeError: If the previous database could not be backed up
-            OSError: If the restored file could not be copied into place
+            ValueError: If the source or its migrated copy does not verify
+            SafetyBackupFailed: If the previous database could not be backed up
+            sqlite3.OperationalError: If SQLite refuses online replacement
         """
-        backup_path = ""
-        if os.path.exists(self.db_path):
-            timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
-            backup_path = f"{self.db_path}.backup_{timestamp}"
-            if not self.backup_database(backup_path):
-                raise RuntimeError("Could not back up the current database")
-            self.logger.info(f"Current database backed up to: {backup_path}")
-
-        self._replace_contents_from(src_path)
-        with PromptModel._init_lock:
-            PromptModel._initialized_paths.discard(self._schema_key())
-        self._ensure_database_exists()
+        staged = self.stage_restore(src_path)
+        try:
+            backup_path = self.publish_staged(staged)
+        finally:
+            staged.discard()
         self.logger.info(f"Database restored from {src_path}")
         return backup_path

database/pair_ingestion.py

@@ -0,0 +1,196 @@
+"""Atomic metadata enrichment for a single known generation context."""
+
+import os
+from typing import Any, Dict
+
+from .models import normalize_image_path
+from .operations import _coerce_file_info
+from .workflow_documents import (
+    TEXT_COLUMNS,
+    DocumentReader,
+    release_documents,
+    resolve_image_row,
+    stored_values,
+)
+
+try:
+    from ..utils.hashing import generate_prompt_hash
+    from ..utils.json_strict import dumps_strict
+    from ..utils.prompt_pair_resolver import RESOLVER_VERSION
+except ImportError:
+    from utils.hashing import generate_prompt_hash
+    from utils.json_strict import dumps_strict
+    from utils.prompt_pair_resolver import RESOLVER_VERSION
+
+
+def ingest_resolved_pair(
+    db: Any, image_path: str, metadata: Dict[str, Any], resolution: Dict[str, Any]
+) -> Dict[str, Any]:
+    state = resolution.get("state", "unresolved")
+    if state != "resolved":
+        return {"status": state, "pairs_added": 0, "links_added": 0}
+    with db._pair_transaction() as conn:
+        return _ingest_resolved_pair_in(conn, db, image_path, metadata, resolution)
+
+
+def _conflict() -> Dict[str, Any]:
+    return {"status": "conflict", "pairs_added": 0, "links_added": 0}
+
+
+def _ingest_resolved_pair_in(
+    conn: Any,
+    db: Any,
+    image_path: str,
+    metadata: Dict[str, Any],
+    resolution: Dict[str, Any],
+    *,
+    provenance: Dict[str, Any] | None = None,
+) -> Dict[str, Any]:
+    """``ingest_resolved_pair`` inside the caller's open ``_pair_transaction``.
+
+    The compatible-only rules are the public ingester's: an image with more
+    than one exact row, another positive, or a prior pair with another
+    negative is a ``conflict`` and nothing is written. The caller owns the
+    transaction, so its own later failure rolls the link back too.
+    ``provenance`` replaces the link's default resolver provenance (Needs
+    review records the user's selection there).
+    """
+    positive, negative = resolution["positive"], resolution["negative"]
+    positive_text = positive.get("storage_text", positive["text"])
+    negative_text = negative.get("storage_text", negative["text"])
+    normalized = normalize_image_path(image_path)
+    if provenance is None:
+        provenance = {
+            "resolver_version": RESOLVER_VERSION,
+            "positive": positive,
+            "negative": negative,
+        }
+    encoded = dumps_strict(provenance, sort_keys=True, separators=(",", ":"))
+    images = conn.execute(
+        "SELECT gi.*,p.hash AS positive_hash FROM generated_images gi "
+        "JOIN prompts p ON p.id=gi.prompt_id "
+        "WHERE gi.file_path=? OR gi.image_path=?",
+        (normalized, image_path),
+    ).fetchall()
+    if len(images) > 1 or (
+        images and images[0]["positive_hash"] != generate_prompt_hash(positive_text)
+    ):
+        return _conflict()
+    existing = images[0] if images else None
+    if existing:
+        prior = conn.execute(
+            "SELECT n.hash AS negative_hash FROM image_prompt_pairs ip "
+            "JOIN prompt_pairs pp ON pp.id=ip.pair_id "
+            "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id WHERE ip.image_id=?",
+            (existing["id"],),
+        ).fetchall()
+        expected = (
+            generate_prompt_hash(negative_text) if negative_text.strip() else None
+        )
+        if any(row["negative_hash"] != expected for row in prior):
+            return _conflict()
+    prior_pair = conn.execute(
+        "SELECT pp.id FROM prompt_pairs pp "
+        "JOIN prompts p ON p.id=pp.positive_prompt_id "
+        "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id "
+        "WHERE p.hash=? AND n.hash IS ?",
+        (
+            generate_prompt_hash(positive_text),
+            generate_prompt_hash(negative_text) if negative_text.strip() else None,
+        ),
+    ).fetchone()
+    pair_id = db._save_pair(conn, positive_text, negative_text)
+    pair = conn.execute("SELECT * FROM prompt_pairs WHERE id=?", (pair_id,)).fetchone()
+    pairs_added = int(prior_pair is None)
+    file_size, width, height, image_format = _coerce_file_info(metadata)
+    # ComfyUI's NaN/Infinity are stored as null: the snapshot is strict JSON.
+    snapshots = {
+        column: dumps_strict(metadata[key], sort_keys=True)
+        for column, key in (
+            ("workflow_data", "workflow"),
+            ("prompt_metadata", "prompt"),
+            ("parameters", "parameters"),
+        )
+        if key in metadata
+    }
+    info = {
+        "file_size": file_size,
+        "width": width,
+        "height": height,
+        "format": image_format,
+    }
+    snapshots.update({key: value for key, value in info.items() if value is not None})
+    changed = False
+    if existing:
+        image_id = existing["id"]
+        # Compared as stored text, whichever form the row is in.
+        current = resolve_image_row(DocumentReader(conn), existing)
+        updates = {
+            key: value for key, value in snapshots.items() if current[key] != value
+        }
+        if updates:
+            replaced = [
+                existing[reference]
+                for column, reference in TEXT_COLUMNS
+                if column in updates
+            ]
+            updates = _document_form(conn, updates)
+            conn.execute(
+                "UPDATE generated_images SET "
+                + ",".join(f"{key}=?" for key in updates)
+                + " WHERE id=?",
+                [*updates.values(), image_id],
+            )
+            release_documents(conn, replaced)
+            changed = True
+    else:
+        stored = _document_form(conn, snapshots)
+        columns = ["prompt_id", "image_path", "filename", "file_path", *stored]
+        values = [
+            pair["positive_prompt_id"],
+            image_path,
+            os.path.basename(image_path),
+            normalized,
+            *stored.values(),
+        ]
+        image_id = conn.execute(
+            "INSERT INTO generated_images("
+            + ",".join(columns)
+            + ") VALUES("
+            + ",".join("?" for _ in columns)
+            + ")",
+            values,
+        ).lastrowid
+    linked = conn.execute(
+        "SELECT 1 FROM image_prompt_pairs WHERE pair_id=? "
+        "AND image_id=? AND provenance=?",
+        (pair_id, image_id, encoded),
+    ).fetchone()
+    db._link_image_to_pair(conn, pair_id, image_id, provenance)
+    status = (
+        "added"
+        if existing is None
+        else "enriched" if changed or not linked else "unchanged"
+    )
+    return {
+        "status": status,
+        "pairs_added": pairs_added,
+        "links_added": int(not linked),
+        "pair_id": pair_id,
+        "image_id": image_id,
+    }
+
+
+def _document_form(conn: Any, values: Dict[str, Any]) -> Dict[str, Any]:
+    """``values`` with the workflow and prompt graph texts stored once.
+
+    Each of those columns becomes NULL beside its document id (new form);
+    every other column is unchanged.
+    """
+    texts = {
+        column: values[column]
+        for column, _reference in TEXT_COLUMNS
+        if column in values
+    }
+    rest = {key: value for key, value in values.items() if key not in texts}
+    return {**rest, **stored_values(conn, texts)}

database/pairs.py

@@ -0,0 +1,433 @@
+"""Ordered reusable prompt pairs and their generation provenance."""
+
+import json
+import sqlite3
+from contextlib import contextmanager
+from typing import Any, Dict, Iterator, List, Optional
+
+from .models import pair_provenance_error, utc_now_iso
+
+try:
+    from ..utils.hashing import generate_prompt_hash
+except ImportError:
+    from utils.hashing import generate_prompt_hash
+
+
+class PairDependencyError(ValueError):
+    """An operation would change or remove content used by a saved pair."""
+
+
+def _validate_metadata(
+    title: Optional[str],
+    category: Optional[str],
+    tags: Optional[List[str]],
+    rating: Optional[int],
+    notes: Optional[str],
+) -> Optional[List[str]]:
+    for name, value in (("title", title), ("category", category), ("notes", notes)):
+        if value is not None and not isinstance(value, str):
+            raise ValueError(f"{name} must be a string")
+    if rating is not None and (type(rating) is not int or not 1 <= rating <= 5):
+        raise ValueError("Rating must be between 1 and 5")
+    if tags is None:
+        return None
+    if not isinstance(tags, list) or any(not isinstance(tag, str) for tag in tags):
+        raise ValueError("tags must be a list of strings")
+    return list(dict.fromkeys(tag.strip() for tag in tags if tag.strip()))
+
+
+class PairOperations:
+    """Connection-level pair operations mixed into PromptDatabase."""
+
+    @contextmanager
+    def _pair_transaction(self) -> Iterator[sqlite3.Connection]:
+        conn = self.model.get_connection()
+        if conn.in_transaction:
+            raise RuntimeError("Pair operations require their own transaction")
+        with conn:
+            conn.execute("BEGIN IMMEDIATE")
+            yield conn
+
+    def _ensure_prompts_unpaired(
+        self, conn: sqlite3.Connection, prompt_ids: List[int]
+    ) -> None:
+        if not prompt_ids:
+            return
+        placeholders = ",".join("?" for _ in prompt_ids)
+        row = conn.execute(
+            f"SELECT id FROM prompt_pairs WHERE positive_prompt_id IN ({placeholders}) "
+            f"OR negative_prompt_id IN ({placeholders}) LIMIT 1",
+            [*prompt_ids, *prompt_ids],
+        ).fetchone()
+        if row:
+            raise PairDependencyError(
+                f"Prompt belongs to saved pair {row['id']}; delete the pair first "
+                "or save changed text as a new prompt"
+            )
+
+    def _pair_component(self, conn: sqlite3.Connection, text: str) -> int:
+        return self._pair_component_outcome(conn, text)[0]
+
+    def _pair_component_outcome(
+        self, conn: sqlite3.Connection, text: str
+    ) -> tuple[int, bool]:
+        """Resolve a component by its dedup hash; report whether it was inserted.
+
+        The first stored spelling wins: the hash ignores outer whitespace and
+        case, internal whitespace stays significant. ``created`` comes from the
+        INSERT result under the caller's held write transaction.
+        """
+        prompt_hash = generate_prompt_hash(text)
+        now = utc_now_iso()
+        created = (
+            conn.execute(
+                "INSERT INTO prompts(text, hash, created_at, updated_at) "
+                "VALUES(?,?,?,?) ON CONFLICT(hash) DO NOTHING",
+                (text.strip(), prompt_hash, now, now),
+            ).rowcount
+            == 1
+        )
+        row = conn.execute(
+            "SELECT id FROM prompts WHERE hash = ?", (prompt_hash,)
+        ).fetchone()
+        return row["id"], created
+
+    def _sync_pair_tags(
+        self, conn: sqlite3.Connection, pair_id: int, tags: List[str]
+    ) -> None:
+        conn.execute("DELETE FROM prompt_pair_tags WHERE pair_id = ?", (pair_id,))
+        for tag_id in self._ensure_tags(conn, tags).values():
+            conn.execute(
+                "INSERT INTO prompt_pair_tags(pair_id,tag_id) VALUES(?,?)",
+                (pair_id, tag_id),
+            )
+
+    def save_prompt_pair(
+        self,
+        positive_text: str,
+        negative_text: str = "",
+        *,
+        title: Optional[str] = None,
+        category: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        rating: Optional[int] = None,
+        notes: Optional[str] = None,
+    ) -> int:
+        """Save an ordered pair atomically; repeat saves preserve its metadata."""
+        with self._pair_transaction() as conn:
+            return self._save_pair(
+                conn,
+                positive_text,
+                negative_text,
+                title=title,
+                category=category,
+                tags=tags,
+                rating=rating,
+                notes=notes,
+            )
+
+    def _save_pair(
+        self,
+        conn: sqlite3.Connection,
+        positive_text: str,
+        negative_text: str = "",
+        *,
+        title: Optional[str] = None,
+        category: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        rating: Optional[int] = None,
+        notes: Optional[str] = None,
+    ) -> int:
+        """Save an ordered pair atomically; repeat saves preserve its metadata."""
+        if not isinstance(positive_text, str) or not positive_text.strip():
+            raise ValueError("Positive prompt text cannot be empty")
+        if not isinstance(negative_text, str):
+            raise ValueError("Negative prompt text must be a string; use '' for empty")
+        # Validate metadata before any component insertion.
+        _validate_metadata(title, category, tags, rating, notes)
+        positive = self._pair_component(conn, positive_text)
+        negative = (
+            self._pair_component(conn, negative_text) if negative_text.strip() else None
+        )
+        return self._save_pair_components(
+            conn,
+            positive,
+            negative,
+            title=title,
+            category=category,
+            tags=tags,
+            rating=rating,
+            notes=notes,
+        )[0]
+
+    def _save_pair_components(
+        self,
+        conn: sqlite3.Connection,
+        positive_id: int,
+        negative_id: Optional[int],
+        *,
+        title: Optional[str] = None,
+        category: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        rating: Optional[int] = None,
+        notes: Optional[str] = None,
+    ) -> tuple[int, bool]:
+        """Create or reuse the ordered pair of existing component IDs.
+
+        Returns ``(pair_id, created)``. A reused pair keeps all of its metadata
+        and archive state; metadata applies only to a newly inserted pair.
+        """
+        tags = _validate_metadata(title, category, tags, rating, notes)
+        existing = conn.execute(
+            "SELECT id FROM prompt_pairs WHERE positive_prompt_id = ? "
+            "AND negative_prompt_id IS ?",
+            (positive_id, negative_id),
+        ).fetchone()
+        if existing:
+            return existing["id"], False
+        now = utc_now_iso()
+        pair_id = conn.execute(
+            "INSERT INTO prompt_pairs(positive_prompt_id,negative_prompt_id,title,"
+            "category,rating,notes,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?)",
+            (positive_id, negative_id, title, category, rating, notes, now, now),
+        ).lastrowid
+        if tags:
+            self._sync_pair_tags(conn, pair_id, tags)
+        return pair_id, True
+
+    def _read_pair_rows(
+        self, conn: sqlite3.Connection, where: str, params: List[Any], suffix: str = ""
+    ) -> List[Dict[str, Any]]:
+        rows = conn.execute(
+            "SELECT pp.*, p.text AS positive_text, "
+            "COALESCE(n.text,'') AS negative_text,"
+            "(SELECT COUNT(DISTINCT image_id) FROM image_prompt_pairs ip "
+            "WHERE ip.pair_id=pp.id) AS image_count FROM prompt_pairs pp "
+            "JOIN prompts p ON p.id=pp.positive_prompt_id "
+            "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id WHERE " + where + suffix,
+            params,
+        ).fetchall()
+        pairs = [dict(row) for row in rows]
+        if not pairs:
+            return pairs
+        by_id = {pair["id"]: pair for pair in pairs}
+        for pair in pairs:
+            pair["archived"] = bool(pair["archived"])
+            pair["tags"] = []
+        placeholders = ",".join("?" for _ in by_id)
+        for row in conn.execute(
+            "SELECT pt.pair_id,t.name FROM prompt_pair_tags pt JOIN tags t ON "
+            f"t.id=pt.tag_id WHERE pt.pair_id IN ({placeholders}) ORDER BY t.name",
+            list(by_id),
+        ):
+            by_id[row["pair_id"]]["tags"].append(row["name"])
+        return pairs
+
+    def get_prompt_pair(self, pair_id: int) -> Optional[Dict[str, Any]]:
+        """Return components, pair metadata and image count, including archives."""
+        rows = self._read_pair_rows(self.model.get_connection(), "pp.id = ?", [pair_id])
+        return rows[0] if rows else None
+
+    def list_prompt_pairs(
+        self,
+        *,
+        text: Optional[str] = None,
+        category: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        include_archived: bool = False,
+        limit: int = 50,
+        offset: int = 0,
+        sort: str = "newest",
+    ) -> Dict[str, Any]:
+        """List pairs with all-tag filtering and stable paginated ordering."""
+        orders = {
+            "newest": "pp.id DESC",
+            "last_used_desc": "julianday(pp.last_used_at) DESC, pp.id DESC",
+        }
+        if sort not in orders:
+            raise ValueError("Unknown pair sort order")
+        if type(limit) is not int or not 1 <= limit <= 500:
+            raise ValueError("limit must be between 1 and 500")
+        if type(offset) is not int or offset < 0:
+            raise ValueError("offset must be a nonnegative integer")
+        tags = _validate_metadata(None, category, tags, None, None)
+        if text is not None and not isinstance(text, str):
+            raise ValueError("text must be a string")
+        clauses, params = ["1=1"], []
+        if not include_archived:
+            clauses.append("pp.archived=0")
+        if category is not None:
+            clauses.append("pp.category = ?")
+            params.append(category)
+        if text:
+            clauses.append("(p.text LIKE ? OR n.text LIKE ? OR pp.title LIKE ?)")
+            params.extend([f"%{text}%"] * 3)
+        for tag in tags or []:
+            clauses.append(
+                "EXISTS(SELECT 1 FROM prompt_pair_tags pt JOIN tags t "
+                "ON t.id=pt.tag_id WHERE pt.pair_id=pp.id AND t.name=?)"
+            )
+            params.append(tag)
+        where = " AND ".join(clauses)
+        conn = self.model.get_connection()
+        total = conn.execute(
+            "SELECT COUNT(*) FROM prompt_pairs pp "
+            "JOIN prompts p ON p.id=pp.positive_prompt_id "
+            "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id WHERE " + where,
+            params,
+        ).fetchone()[0]
+        pairs = self._read_pair_rows(
+            conn,
+            where,
+            [*params, limit, offset],
+            f" ORDER BY {orders[sort]} LIMIT ? OFFSET ?",
+        )
+        return {
+            "pairs": pairs,
+            "total": total,
+            "limit": limit,
+            "offset": offset,
+            "has_more": offset + len(pairs) < total,
+        }
+
+    def update_prompt_pair_metadata(
+        self,
+        pair_id: int,
+        *,
+        title: Optional[str] = None,
+        category: Optional[str] = None,
+        tags: Optional[List[str]] = None,
+        rating: Optional[int] = None,
+        notes: Optional[str] = None,
+    ) -> bool:
+        """Update pair metadata explicitly; None preserves and empty values clear."""
+        tags = _validate_metadata(title, category, tags, rating, notes)
+        changes = {
+            key: value
+            for key, value in (
+                ("title", title),
+                ("category", category),
+                ("rating", rating),
+                ("notes", notes),
+            )
+            if value is not None
+        }
+        if not changes and tags is None:
+            return False
+        with self._pair_transaction() as conn:
+            if not conn.execute(
+                "SELECT 1 FROM prompt_pairs WHERE id=?", (pair_id,)
+            ).fetchone():
+                return False
+            changes["updated_at"] = utc_now_iso()
+            conn.execute(
+                "UPDATE prompt_pairs SET "
+                + ",".join(f"{key}=?" for key in changes)
+                + " WHERE id=?",
+                [*changes.values(), pair_id],
+            )
+            if tags is not None:
+                self._sync_pair_tags(conn, pair_id, tags)
+            return True
+
+    def archive_prompt_pair(self, pair_id: int, archived: bool = True) -> bool:
+        """Hide a reusable pair without changing components or generation history."""
+        if type(archived) is not bool:
+            raise ValueError("archived must be a boolean")
+        with self._pair_transaction() as conn:
+            return (
+                conn.execute(
+                    "UPDATE prompt_pairs SET archived=?,updated_at=? WHERE id=?",
+                    (int(archived), utc_now_iso(), pair_id),
+                ).rowcount
+                > 0
+            )
+
+    def delete_prompt_pair(self, pair_id: int) -> bool:
+        """Delete unused pair metadata; generated pairs must be archived instead."""
+        with self._pair_transaction() as conn:
+            if conn.execute(
+                "SELECT 1 FROM image_prompt_pairs WHERE pair_id=? LIMIT 1", (pair_id,)
+            ).fetchone():
+                raise PairDependencyError(
+                    "Pair has generated images; archive it instead"
+                )
+            return (
+                conn.execute("DELETE FROM prompt_pairs WHERE id=?", (pair_id,)).rowcount
+                > 0
+            )
+
+    def record_prompt_pair_use(self, pair_id: int, times: int = 1) -> bool:
+        """Count pair uses independently of standalone prompt usage."""
+        if type(times) is not int or times < 1:
+            raise ValueError("times must be a positive integer")
+        with self._pair_transaction() as conn:
+            return (
+                conn.execute(
+                    "UPDATE prompt_pairs SET run_count=run_count+?,last_used_at=? "
+                    "WHERE id=?",
+                    (times, utc_now_iso(), pair_id),
+                ).rowcount
+                > 0
+            )
+
+    def link_image_to_pair(
+        self, pair_id: int, image_id: int, provenance: Dict[str, Any]
+    ) -> int:
+        """Associate an image with immutable run provenance, idempotently."""
+        with self._pair_transaction() as conn:
+            return self._link_image_to_pair(conn, pair_id, image_id, provenance)
+
+    def _link_image_to_pair(
+        self,
+        conn: sqlite3.Connection,
+        pair_id: int,
+        image_id: int,
+        provenance: Dict[str, Any],
+    ) -> int:
+        """Associate an image with immutable run provenance, idempotently."""
+        if not isinstance(provenance, dict):
+            raise ValueError("provenance must be an object")
+        try:
+            encoded = json.dumps(
+                provenance, sort_keys=True, separators=(",", ":"), allow_nan=False
+            )
+        except (TypeError, ValueError, RecursionError) as error:
+            raise ValueError("provenance must contain JSON values") from error
+        error = pair_provenance_error(encoded)
+        if error:
+            raise ValueError(error)
+        pair = conn.execute(
+            "SELECT positive_prompt_id FROM prompt_pairs WHERE id=?", (pair_id,)
+        ).fetchone()
+        image = conn.execute(
+            "SELECT prompt_id FROM generated_images WHERE id=?", (image_id,)
+        ).fetchone()
+        if not pair or not image:
+            raise ValueError("Pair and image must exist")
+        if pair["positive_prompt_id"] != image["prompt_id"]:
+            raise ValueError("Image must belong to the pair's positive prompt")
+        conn.execute(
+            "INSERT INTO image_prompt_pairs(pair_id,image_id,provenance,created_at) "
+            "VALUES(?,?,?,?) ON CONFLICT(pair_id,image_id,provenance) DO NOTHING",
+            (pair_id, image_id, encoded, utc_now_iso()),
+        )
+        return conn.execute(
+            "SELECT id FROM image_prompt_pairs "
+            "WHERE pair_id=? AND image_id=? AND provenance=?",
+            (pair_id, image_id, encoded),
+        ).fetchone()["id"]
+
+    def get_image_prompt_pairs(self, image_id: int) -> List[Dict[str, Any]]:
+        """Return every recorded pair association with decoded provenance."""
+        rows = (
+            self.model.get_connection()
+            .execute(
+                "SELECT * FROM image_prompt_pairs WHERE image_id=? ORDER BY id",
+                (image_id,),
+            )
+            .fetchall()
+        )
+        return [
+            {**dict(row), "provenance": json.loads(row["provenance"])} for row in rows
+        ]

database/restore_stage.py

@@ -0,0 +1,365 @@
+"""A restore in two steps: stage the upload, then publish it (#249 System S8).
+
+**Stage** (``stage_restore``) never touches the live database or the source.
+In a new folder ``.pm-restore-*`` in the library's own folder (#318: never
+the system temp folder, which may be another disk), it:
+
+1. checks the source (``verify_database_file`` without reading every
+   workflow document: a damaged one is repaired below, not refused);
+2. copies it with SQLite's backup API (a rollback-journal file and a WAL file
+   alike) and opens the copy as a ``PromptModel``, which migrates an older
+   schema;
+3. repairs damaged workflow documents (``repair_damaged_documents``),
+   rotates the history epoch and counts what the copy holds;
+4. makes the copy one self-contained file and verifies it fully (every
+   current table, every document);
+5. records the copy's SHA-256, so publishing binds exactly these bytes.
+
+**Publish** (``publish_staged``) re-hashes the stage (``StageChanged`` if it
+differs), keeps the current database as a uniquely named, standalone safety
+backup beside it (``database/safety_backup.py``; not verified, because the
+database being replaced may be the damaged one), copies the stage over the
+live pages through the backup API (no file is swapped under open handles; an
+``OperationalError`` there leaves the live data intact, #226), then
+checkpoints the WAL with TRUNCATE so the copied pages do not leave a
+database-sized ``-wal`` behind. A reader that keeps a snapshot past the busy
+wait is logged, not fatal: the restore itself is done.
+
+Every handle on the stage is closed before its folder is removed (Windows).
+``sweep_restore_temps`` removes what an interrupted restore left (startup).
+"""
+
+import hashlib
+import logging
+import os
+import sqlite3
+import tempfile
+from contextlib import closing
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Dict, Optional
+
+from . import workflow_documents as wd
+from .execution_history_schema import rotate_history_epoch
+from .safety_backup import take_safety_backup
+
+logger = logging.getLogger("prompt_manager.database.restore_stage")
+
+STAGE_PREFIX = ".pm-restore-"
+UPLOAD_PREFIX = ".pm-restore-upload-"
+STAGED_NAME = "staged.db"
+# How long the post-restore TRUNCATE checkpoint waits for readers to leave.
+CHECKPOINT_BUSY_MS = 30_000
+_HASH_CHUNK = 1024 * 1024
+SUMMARY_KEYS = (
+    "prompts",
+    "positives",
+    "pairs",
+    "archived_pairs",
+    "image_links",
+    "runs",
+    "saved_views",
+)
+
+
+class StageChanged(ValueError):
+    """The staged file is no longer the bytes that were reviewed."""
+
+
+@dataclass
+class StagedRestore:
+    """A verified, migrated copy of an upload, ready to publish."""
+
+    path: str
+    digest: str
+    size: int
+    summary: Dict[str, int]
+    schema: Dict[str, bool]
+    documents: Dict[str, int] = field(
+        default_factory=lambda: {"repaired": 0, "cleared": 0, "images": 0}
+    )
+
+    @property
+    def directory(self) -> str:
+        return os.path.dirname(self.path)
+
+    def discard(self) -> None:
+        """Remove the stage folder (idempotent; its handles are all closed)."""
+        remove_stage_folder(self.directory)
+
+
+def remove_stage_folder(directory: str) -> None:
+    """Remove a stage folder and its files, logging what cannot be removed."""
+    if not directory or not os.path.isdir(directory):
+        return
+    for name in os.listdir(directory):
+        try:
+            os.remove(os.path.join(directory, name))
+        except OSError as error:
+            logger.warning(f"Could not remove restore stage file {name}: {error}")
+    try:
+        os.rmdir(directory)
+    except OSError as error:
+        logger.warning(f"Could not remove the restore stage folder: {error}")
+
+
+def file_digest(path: str) -> str:
+    """``sha256:<hex>`` of the file's bytes (blocking)."""
+    digest = hashlib.sha256()
+    with open(path, "rb") as handle:
+        for chunk in iter(lambda: handle.read(_HASH_CHUNK), b""):
+            digest.update(chunk)
+    return "sha256:" + digest.hexdigest()
+
+
+def _readonly(path: str) -> sqlite3.Connection:
+    return sqlite3.connect(Path(path).resolve().as_uri() + "?mode=ro", uri=True)
+
+
+def _schema(path: str) -> list:
+    """The file's schema objects, to tell whether staging migrated it."""
+    with closing(_readonly(path)) as conn:
+        return sorted(
+            conn.execute(
+                "SELECT type, name, tbl_name, sql FROM sqlite_master "
+                "WHERE name NOT LIKE 'sqlite_%'"
+            ).fetchall(),
+            key=repr,
+        )
+
+
+def _count(conn: sqlite3.Connection, table: str, where: str = "") -> int:
+    exists = conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type='table' AND name=?", (table,)
+    ).fetchone()
+    if not exists:
+        return 0
+    # ``table`` and ``where`` are constants of library_summary, never input.
+    return conn.execute(f"SELECT COUNT(*) FROM {table} {where}").fetchone()[0]  # nosec
+
+
+def library_summary(conn: sqlite3.Connection) -> Dict[str, int]:
+    """What a library holds, for a restore's review (reads only).
+
+    ``positives`` are the Library's prompts (``library_counts``'s
+    ``positive_prompts``); ``pairs`` the active saved pairs and
+    ``archived_pairs`` the archived ones; ``image_links`` the indexed image
+    rows; ``runs`` the recorded executions.
+    """
+    from .library import _ELIGIBLE
+
+    has_state = conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type='table' AND "
+        "name='prompt_library_state'"
+    ).fetchone()
+    has_pairs = conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type='table' AND name='prompt_pairs'"
+    ).fetchone()
+    if has_state and has_pairs:
+        positives = conn.execute(
+            "SELECT COUNT(*) FROM prompts p WHERE " + _ELIGIBLE
+        ).fetchone()[0]
+    else:
+        positives = _count(conn, "prompts")
+    return {
+        "prompts": _count(conn, "prompts"),
+        "positives": positives,
+        "pairs": _count(conn, "prompt_pairs", "WHERE archived=0"),
+        "archived_pairs": _count(conn, "prompt_pairs", "WHERE archived<>0"),
+        "image_links": _count(conn, "generated_images"),
+        "runs": _count(conn, "execution_runs"),
+        "saved_views": _count(conn, "saved_views"),
+    }
+
+
+# The newest row of each table a restore would overwrite: with the counts,
+# a write after the review (a new run, image, prompt or edit) changes it.
+_FINGERPRINT_SQL = (
+    ("prompts", "SELECT MAX(id), MAX(updated_at), MAX(last_used_at) FROM prompts"),
+    ("generated_images", "SELECT MAX(id) FROM generated_images"),
+    ("prompt_pairs", "SELECT MAX(id), MAX(updated_at) FROM prompt_pairs"),
+    ("execution_runs", "SELECT COUNT(*), MAX(rowid) FROM execution_runs"),
+    ("saved_views", "SELECT MAX(rowid) FROM saved_views"),
+)
+
+
+def live_fingerprint(conn: sqlite3.Connection):
+    """``(summary, fingerprint)`` of a live library, read in one snapshot.
+
+    The fingerprint changes when rows are added to or removed from the
+    tables a restore replaces, or when a prompt, pair, run or saved view is
+    the newest edited (counts plus the newest row and edit of each table);
+    the apply refuses with ``plan-changed`` then. It does not see every
+    edit (a tag or favorite change on an older row); those are kept in the
+    restore's safety backup. A read transaction the caller already has is
+    used and left open; one opened here is ended here.
+    """
+    import json
+
+    opened = not conn.in_transaction
+    if opened:
+        conn.execute("BEGIN")  # one read snapshot for every count
+    try:
+        summary = library_summary(conn)
+        newest = []
+        for table, sql in _FINGERPRINT_SQL:
+            try:
+                newest.append([table, list(conn.execute(sql).fetchone())])
+            except sqlite3.OperationalError:
+                newest.append([table, None])  # no such table or column
+    finally:
+        if opened:
+            conn.rollback()  # it only read
+    text = json.dumps([summary, newest], sort_keys=True, default=str)
+    return summary, "sha256:" + hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def _copy(src_path: str, staged_path: str) -> None:
+    source = _readonly(src_path)
+    try:
+        with closing(sqlite3.connect(staged_path)) as destination:
+            source.backup(destination)
+    finally:
+        source.close()
+
+
+def _standalone(path: str) -> None:
+    with closing(sqlite3.connect(path)) as conn:
+        conn.execute("PRAGMA journal_mode=DELETE").fetchone()
+
+
+def stage_restore(model_cls, db_path: str, src_path: str) -> StagedRestore:
+    """Stage ``src_path`` beside ``db_path`` (blocking); see the module docstring.
+
+    Raises ValueError with a readable reason when the source or the migrated
+    stage does not verify; the stage folder is then already removed.
+    """
+    ok, reason = model_cls.verify_database_file(src_path, documents=False)
+    if not ok:
+        raise ValueError(reason)
+    before = _schema(src_path)
+    with closing(_readonly(src_path)) as conn:
+        library_state = bool(
+            conn.execute(
+                "SELECT 1 FROM sqlite_master WHERE type='table' AND "
+                "name='prompt_library_state'"
+            ).fetchone()
+        )
+    folder = os.path.dirname(os.path.abspath(db_path))
+    directory = tempfile.mkdtemp(prefix=STAGE_PREFIX, dir=folder)
+    staged_path = os.path.join(directory, STAGED_NAME)
+    staged_key = os.path.normcase(os.path.realpath(staged_path))
+    try:
+        _copy(src_path, staged_path)
+        staged = None
+        try:
+            staged = model_cls(staged_path)
+            conn = staged.get_connection()
+            documents = wd.repair_damaged_documents(conn)
+            rotate_history_epoch(conn)
+            summary = library_summary(conn)
+        finally:
+            if staged is not None:
+                staged.close_all()
+        _standalone(staged_path)
+        ok, reason = model_cls.verify_database_file(
+            staged_path, require_library_state=True
+        )
+        if not ok:
+            raise ValueError(reason)
+        result = StagedRestore(
+            path=staged_path,
+            digest=file_digest(staged_path),
+            size=os.path.getsize(staged_path),
+            summary=summary,
+            schema={
+                "migrated": _schema(staged_path) != before,
+                "library_state": library_state,
+            },
+            documents=documents,
+        )
+    except BaseException:
+        remove_stage_folder(directory)
+        raise
+    finally:
+        with model_cls._init_lock:
+            model_cls._initialized_paths.discard(staged_key)
+    return result
+
+
+def checkpoint_truncate(conn: sqlite3.Connection) -> bool:
+    """Checkpoint the WAL and truncate it to 0 bytes; False when a reader held it."""
+    previous = conn.execute("PRAGMA busy_timeout").fetchone()[0]
+    try:
+        conn.execute(f"PRAGMA busy_timeout = {int(CHECKPOINT_BUSY_MS)}")
+        busy = conn.execute("PRAGMA wal_checkpoint(TRUNCATE)").fetchone()[0]
+    finally:
+        conn.execute(f"PRAGMA busy_timeout = {int(previous)}")
+    return not busy
+
+
+def publish_staged(model, staged: StagedRestore) -> str:
+    """Publish ``staged`` over ``model``'s database; return the safety backup path.
+
+    "" when there was no database to back up. Raises StageChanged before
+    anything is written when the stage's bytes changed since staging, and
+    SafetyBackupFailed when the current database could not be kept. An
+    ``OperationalError`` from the page copy propagates with
+    ``_prompt_manager_restore_backup_created`` set; the live data is intact.
+    """
+    if not os.path.isfile(staged.path) or file_digest(staged.path) != staged.digest:
+        raise StageChanged("The staged restore changed since it was reviewed")
+    backup_path = ""
+    if os.path.exists(model.db_path):
+        backup_path = take_safety_backup(model, verify=False)
+        logger.info(f"Current database backed up to: {os.path.basename(backup_path)}")
+    try:
+        model._replace_contents_from(staged.path)
+    except sqlite3.OperationalError as exc:
+        exc._prompt_manager_restore_backup_created = bool(backup_path)
+        raise
+    with model._init_lock:
+        model._initialized_paths.add(model._schema_key())
+    _truncate_wal(model)
+    return backup_path
+
+
+def _truncate_wal(model) -> Optional[bool]:
+    """TRUNCATE-checkpoint the live WAL after a publish (never raises)."""
+    try:
+        conn = model.get_connection()
+        with conn.using():
+            truncated = checkpoint_truncate(conn)
+    except sqlite3.Error as error:
+        logger.warning(f"The database log could not be truncated: {error}")
+        return None
+    if not truncated:
+        logger.warning(
+            "The database log could not be truncated after the restore: a "
+            "reader kept it open. It shrinks at the next checkpoint with no "
+            "reader."
+        )
+    return truncated
+
+
+def sweep_restore_temps(db_path: str) -> list:
+    """Remove the stages and uploads an interrupted restore left (startup)."""
+    folder = os.path.dirname(os.path.abspath(db_path))
+    removed = []
+    try:
+        names = os.listdir(folder)
+    except OSError:
+        return removed
+    for name in names:
+        path = os.path.join(folder, name)
+        if name.startswith(UPLOAD_PREFIX) and os.path.isfile(path):
+            try:
+                os.remove(path)
+                removed.append(name)
+            except OSError:
+                continue
+        elif name.startswith(STAGE_PREFIX) and os.path.isdir(path):
+            remove_stage_folder(path)
+            if not os.path.exists(path):
+                removed.append(name)
+    return removed

database/review.py

@@ -0,0 +1,648 @@
+"""Needs review items (#249): persist and read capture-bound ambiguous evidence.
+
+An item is one ambiguous resolver result for one annotation byte record (the
+whole-file SHA-256 of one physical file at one locator), keyed by
+``(byte_id, resolver_version, evidence_hash)``. Recording is idempotent: the
+same evidence for the same bytes is a duplicate and never touches an existing
+item, its state, revision or decisions. Changed bytes are a new byte record
+and therefore a new pending item; the old one stays.
+
+Callers pass verified facts (the capture's digest, size and ``mtime_ns``);
+nothing here reads the filesystem. Writes run in one ``_pair_transaction``
+fenced by the storage epoch the capture was taken under.
+"""
+
+import hashlib
+import json
+import os
+import re
+import sqlite3
+from typing import Iterator
+
+try:
+    from ..utils.hashing import generate_prompt_hash
+except ImportError:
+    from utils.hashing import generate_prompt_hash
+
+from .annotation_schema import (
+    BYTES_TABLE,
+    FILES_TABLE,
+    METADATA_TABLE as ANNOTATION_METADATA_TABLE,
+)
+from .annotations import _new_id, _validate_locator, _validate_target
+from .models import normalize_image_path, utc_now_iso
+from .review_changes import notify_review_change
+from .review_schema import (
+    DECISIONS_TABLE,
+    ITEMS_TABLE,
+    MAX_CANDIDATES,
+    MAX_SPELLED_PATH_BYTES,
+    METADATA_TABLE,
+    canonical_evidence,
+    evidence_digest,
+    evidence_offers_choice,
+)
+
+# Only this module's table constants are interpolated into the SQL below.
+_INSERT_ITEM = (
+    f"INSERT OR IGNORE INTO {ITEMS_TABLE}(id,byte_id,spelled_path,"  # nosec B608
+    "resolver_version,evidence_hash,evidence,candidate_count,limited,state,"
+    "revision,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?,'pending',0,?,?)"
+)
+_BUMP_REVIEW_INDEX = (
+    f"UPDATE {METADATA_TABLE} SET "  # nosec B608
+    "review_index_revision=review_index_revision+1 WHERE singleton_id=1"
+)
+# A later spelling of the same locator (a case-only rename on a case-folding
+# filesystem) replaces the recorded one, so targets keep formatting.
+_REFRESH_SPELLING = (
+    f"UPDATE {ITEMS_TABLE} SET spelled_path=? "  # nosec B608
+    f"WHERE spelled_path<>? AND byte_id IN (SELECT b.id FROM {BYTES_TABLE} b "
+    f"JOIN {FILES_TABLE} f ON f.id=b.file_id WHERE f.locator=?)"
+)
+_REVIEW_INDEX = (
+    f"SELECT review_index_revision FROM {METADATA_TABLE} "  # nosec B608
+    "WHERE singleton_id=1"
+)
+_BUMP_IMAGE_INDEX = (
+    f"UPDATE {ANNOTATION_METADATA_TABLE} SET "  # nosec B608
+    "image_index_revision=image_index_revision+1 WHERE singleton_id=1"
+)
+_ITEMS_OF_LOCATOR = (
+    "SELECT i.id,i.byte_id,i.resolver_version,i.evidence_hash,"  # nosec B608
+    "i.evidence,i.candidate_count,i.limited,i.state,i.revision,i.created_at,"
+    f"i.updated_at,b.sha256 FROM {FILES_TABLE} f "
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? ORDER BY i.created_at,i.id"
+)
+_QUEUED = (
+    f"SELECT 1 FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND b.size=? AND b.mtime_ns=? AND i.resolver_version=? "
+    "LIMIT 1"
+)
+_PENDING_AT = (
+    f"SELECT 1 FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND b.size=? AND b.mtime_ns=? AND i.resolver_version=? "
+    "AND i.state='pending' LIMIT 1"
+)
+# Bytes verified at this stat whose item carries a decision (any version).
+_DECIDED_AT = (
+    f"SELECT 1 FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND b.size=? AND b.mtime_ns=? "
+    "AND (i.state<>'pending' OR i.revision<>0 OR EXISTS (SELECT 1 FROM "
+    f"{DECISIONS_TABLE} d WHERE d.item_id=i.id)) LIMIT 1"
+)
+# The other items of one byte record (another version or evidence), with
+# whether each may be superseded: pending, revision 0, no decision row.
+_OTHER_ITEMS_OF_BYTES = (
+    "SELECT i.id, (i.state='pending' AND i.revision=0 AND NOT EXISTS "  # nosec B608
+    f"(SELECT 1 FROM {DECISIONS_TABLE} d WHERE d.item_id=i.id)) "
+    f"FROM {ITEMS_TABLE} i WHERE i.byte_id=? "
+    "AND NOT (i.resolver_version=? AND i.evidence_hash=?)"
+)
+# Pending items of the bytes verified at one stat, with their versions.
+_PENDING_AT_STAT = (
+    f"SELECT i.id,i.resolver_version FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND b.size=? AND b.mtime_ns=? AND i.state='pending'"
+)
+# Pending items of one file: the candidates a link settles.
+_PENDING_OF_LOCATOR = (
+    f"SELECT i.id FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND i.state='pending'"
+)
+# Pending items of one file with their evidence: an untraced file's settle.
+_PENDING_EVIDENCE_OF_LOCATOR = (
+    f"SELECT i.id,i.evidence FROM {FILES_TABLE} f "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.file_id=f.id "
+    f"JOIN {ITEMS_TABLE} i ON i.byte_id=b.id "
+    "WHERE f.locator=? AND i.state='pending'"
+)
+# Only an undecided item goes: a decision row (RESTRICT) is never deleted.
+_SETTLE_ITEM = (
+    f"DELETE FROM {ITEMS_TABLE} WHERE id=? AND state='pending' "  # nosec B608
+    f"AND revision=0 AND NOT EXISTS (SELECT 1 FROM {DECISIONS_TABLE} d "
+    f"WHERE d.item_id={ITEMS_TABLE}.id)"
+)
+_EPOCH = "SELECT storage_epoch FROM execution_history_metadata WHERE singleton_id=1"
+_PENDING_COUNT = (
+    f"SELECT COUNT(*) FROM {ITEMS_TABLE} WHERE state='pending'"  # nosec B608
+)
+_QUEUE_COLUMNS = (
+    "id",
+    "state",
+    "revision",
+    "created_at",
+    "candidate_count",
+    "limited",
+    "spelled_path",
+    "locator",
+)
+_QUEUE_SELECT = (
+    "SELECT i.id,i.state,i.revision,i.created_at,i.candidate_count,"  # nosec B608
+    f"i.limited,i.spelled_path,f.locator FROM {ITEMS_TABLE} i "
+    f"JOIN {BYTES_TABLE} b ON b.id=i.byte_id "
+    f"JOIN {FILES_TABLE} f ON f.id=b.file_id "
+)
+_QUEUE_FIRST = _QUEUE_SELECT + (
+    "WHERE i.state='pending' ORDER BY i.created_at,i.id LIMIT ?"
+)
+_QUEUE_AFTER = _QUEUE_SELECT + (
+    "WHERE i.state='pending' AND (i.created_at,i.id)>(?,?) "
+    "ORDER BY i.created_at,i.id LIMIT ?"
+)
+_DETAIL_COLUMNS = _QUEUE_COLUMNS + (
+    "byte_id",
+    "resolver_version",
+    "evidence_hash",
+    "evidence",
+    "updated_at",
+    "sha256",
+)
+_DETAIL = (
+    "SELECT i.id,i.state,i.revision,i.created_at,i.candidate_count,"  # nosec B608
+    "i.limited,i.spelled_path,f.locator,i.byte_id,i.resolver_version,"
+    "i.evidence_hash,i.evidence,i.updated_at,b.sha256 "
+    f"FROM {ITEMS_TABLE} i JOIN {BYTES_TABLE} b ON b.id=i.byte_id "
+    f"JOIN {FILES_TABLE} f ON f.id=b.file_id WHERE i.id=?"
+)
+# The legacy rows and pair links the ingester's compatibility rules read.
+_DEPENDENT_IMAGES = (
+    "SELECT gi.id,p.hash,gi.image_path FROM generated_images gi "
+    "JOIN prompts p ON p.id=gi.prompt_id "
+    "WHERE gi.file_path=? OR gi.image_path=? ORDER BY gi.id"
+)
+_DEPENDENT_LINKS = (
+    "SELECT DISTINCT ip.pair_id,n.hash FROM image_prompt_pairs ip "
+    "JOIN prompt_pairs pp ON pp.id=ip.pair_id "
+    "LEFT JOIN prompts n ON n.id=pp.negative_prompt_id "
+    "WHERE ip.image_id=? ORDER BY ip.pair_id"
+)
+# Rows whose spelling ends with the file's name: the candidates for another
+# spelling of the same physical file (the link summaries' filename fallback).
+_SAME_NAME_ROWS = (
+    "SELECT id,image_path FROM generated_images "
+    "WHERE image_path LIKE ? ESCAPE '\\' ORDER BY id"
+)
+_TITLES = (
+    "SELECT p.hash,s.title FROM prompts p "
+    "JOIN prompt_library_state s ON s.prompt_id=p.id "
+    "WHERE p.hash IN ({marks}) AND s.title IS NOT NULL ORDER BY p.id"
+)
+_ITEM_ID = re.compile(r"[0-9a-f]{8}(?:-[0-9a-f]{4}){3}-[0-9a-f]{12}")
+_ITEM_COLUMNS = (
+    "id",
+    "byte_id",
+    "resolver_version",
+    "evidence_hash",
+    "evidence",
+    "candidate_count",
+    "limited",
+    "state",
+    "revision",
+    "created_at",
+    "updated_at",
+    "sha256",
+)
+
+
+def _validate_spelling(spelled_path: object, locator: str) -> None:
+    if (
+        not isinstance(spelled_path, str)
+        or not spelled_path
+        or len(spelled_path.encode("utf-8", "surrogatepass")) > MAX_SPELLED_PATH_BYTES
+        or os.path.normcase(spelled_path) != locator
+    ):
+        raise ValueError("spelled_path must be the file's spelling of its locator")
+
+
+def dependent_images(conn: sqlite3.Connection, spellings) -> list:
+    """``(id, image_path, positive_hash)`` of every legacy row of the file.
+
+    ``spellings`` are the paths the file is linked under (the Apply path and
+    every alias of the same physical file); each is matched as the ingester
+    matches one path. Rows are distinct and ordered by id.
+    """
+    found = {}
+    for spelled in dict.fromkeys(spellings):
+        for row in conn.execute(
+            _DEPENDENT_IMAGES, (normalize_image_path(spelled), spelled)
+        ):
+            found[row[0]] = (row[0], row[2], row[1])
+    return [found[key] for key in sorted(found)]
+
+
+def same_name_rows(conn: sqlite3.Connection, name: str) -> list:
+    """``(id, image_path)`` of rows whose spelling may end with ``name``."""
+    # Imported here: operations composes this module's mixin.
+    from .operations import escape_like
+
+    return [
+        (row[0], row[1])
+        for row in conn.execute(_SAME_NAME_ROWS, (f"%{escape_like(name)}",))
+    ]
+
+
+def dependency_fingerprint(
+    conn: sqlite3.Connection, image_path: str, aliases=()
+) -> str:
+    """SHA-256 of what Apply's compatibility depends on for ``image_path``.
+
+    The exact legacy rows of the file under ``image_path`` and every alias
+    spelling of it, with their positive hashes, and each row's pair links
+    with the negative hash, read as the ingester reads them. Unrelated prompt
+    changes (tags, titles) do not move it.
+    """
+    rows = []
+    for image_id, _spelled, positive_hash in dependent_images(
+        conn, (image_path, *aliases)
+    ):
+        links = conn.execute(_DEPENDENT_LINKS, (image_id,)).fetchall()
+        rows.append([image_id, positive_hash, [list(link) for link in links]])
+    text = canonical_evidence({"v": 1, "images": rows})
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+class ReviewOperations:
+    """PromptDatabase mixin: insert and read Needs review items."""
+
+    def record_review_item(
+        self,
+        *,
+        locator: str,
+        spelled_path: str,
+        sha256: str,
+        size: int,
+        mtime_ns: int,
+        epoch: str,
+        resolver_version: int,
+        evidence: dict,
+        candidate_count: int,
+        limited: bool,
+    ) -> bool:
+        """Store one ambiguous result for these exact bytes; True when new.
+
+        The byte record is created (or its verified fingerprint refreshed)
+        like an observe. ``spelled_path`` is the file's own spelling (it must
+        fold to ``locator``); targets are formatted from it. Raises
+        ``LibraryCommandError("storage-changed")`` when the storage epoch is
+        no longer ``epoch`` (a restore since the capture) and ``ValueError``
+        for invalid facts or evidence.
+
+        One file never holds two pending items for the same bytes: the new
+        item supersedes the byte record's other items (an older resolver or
+        semantics version) when every one is pending and undecided (the safe
+        delete of the self-heal); when one carries a decision or a later
+        revision, it is kept and nothing new is stored (False).
+        """
+        _validate_target(locator, sha256, size, mtime_ns)
+        _validate_spelling(spelled_path, locator)
+        if type(resolver_version) is not int or resolver_version < 1:
+            raise ValueError("resolver_version must be a positive integer")
+        if (
+            type(candidate_count) is not int
+            or not 0 <= candidate_count <= MAX_CANDIDATES
+        ):
+            raise ValueError(f"candidate_count must be between 0 and {MAX_CANDIDATES}")
+        if not isinstance(limited, bool) or (limited and candidate_count):
+            raise ValueError("A limited item carries no candidates")
+        text = canonical_evidence(evidence)
+        evidence_hash = evidence_digest(text)
+        with self._pair_transaction() as conn:
+            self._library_epoch(conn, epoch)
+            now = utc_now_iso()
+            file_id = self._annotation_owner(conn, locator, now)
+            byte_id, _revision, created, moved = self._annotation_byte(
+                conn, file_id, sha256, size, mtime_ns, now
+            )
+            if created or moved:
+                conn.execute(_BUMP_IMAGE_INDEX)
+            # The byte record is created or refreshed even when the item is
+            # refused below (an observe would do the same); that commits
+            # with the image index bump, and the review queue is unchanged.
+            others = conn.execute(
+                _OTHER_ITEMS_OF_BYTES, (byte_id, resolver_version, evidence_hash)
+            ).fetchall()
+            if any(not removable for _id, removable in others):
+                return False
+            superseded = sum(
+                conn.execute(_SETTLE_ITEM, (item_id,)).rowcount
+                for item_id, _removable in others
+            )
+            cursor = conn.execute(
+                _INSERT_ITEM,
+                (
+                    _new_id(),
+                    byte_id,
+                    spelled_path,
+                    resolver_version,
+                    evidence_hash,
+                    text,
+                    candidate_count,
+                    int(limited),
+                    now,
+                    now,
+                ),
+            )
+            inserted = cursor.rowcount > 0
+            refreshed = _refresh_spelling(conn, locator, spelled_path)
+            if inserted or refreshed or superseded:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        if inserted or refreshed or superseded:
+            notify_review_change(self.model.db_path)
+        return inserted
+
+    def refresh_review_spelling(self, locator: str, spelled_path: str) -> bool:
+        """Record ``spelled_path`` as the spelling of every item at ``locator``.
+
+        The rescan's skip calls it for a file it does not copy again: a
+        case-only rename keeps size and ``mtime_ns``, and a target formatted
+        from the old spelling would be refused. True when a spelling changed
+        (the review index revision moves; item revisions do not, so an open
+        decision stays valid).
+        """
+        _validate_locator(locator)
+        _validate_spelling(spelled_path, locator)
+        with self._pair_transaction() as conn:
+            refreshed = _refresh_spelling(conn, locator, spelled_path)
+            if refreshed:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        if refreshed:
+            notify_review_change(self.model.db_path)
+        return refreshed
+
+    def review_index_revision(self) -> int:
+        """The queue's revision (moved by every queue change)."""
+        row = self.model.get_connection().execute(_REVIEW_INDEX).fetchone()
+        if row is None:
+            raise sqlite3.DatabaseError("Missing review queue metadata")
+        return row[0]
+
+    def review_queue_version(self) -> dict:
+        """``{library_epoch, review_revision}`` of the queue, from one snapshot.
+
+        The pair a list read answers: equal pairs mean the same queue (a
+        restore moves the epoch, every queue change moves the revision).
+        """
+        with self._annotation_read() as conn:
+            epoch = conn.execute(_EPOCH).fetchone()
+            revision = conn.execute(_REVIEW_INDEX).fetchone()
+        if epoch is None or revision is None:
+            raise sqlite3.DatabaseError("Missing review queue metadata")
+        return {"library_epoch": epoch[0], "review_revision": revision[0]}
+
+    def review_items_for(self, locator: str) -> list:
+        """Every item of the file at ``locator``, oldest first, evidence parsed."""
+        _validate_locator(locator)
+        return list(self._iter_review_items(locator))
+
+    def _iter_review_items(self, locator: str) -> Iterator[dict]:
+        conn = self.model.get_connection()
+        for row in conn.execute(_ITEMS_OF_LOCATOR, (locator,)):
+            item = dict(zip(_ITEM_COLUMNS, tuple(row)))
+            item["evidence"] = json.loads(item["evidence"])
+            item["limited"] = bool(item["limited"])
+            yield item
+
+    # -- queue reads (#249 E3) -------------------------------------------------
+
+    def review_fingerprint_queued(
+        self, locator: str, size: int, mtime_ns: int, resolver_version: int
+    ) -> bool:
+        """True when an item cites a byte record verified at this exact stat.
+
+        A rescan uses it to skip the copy and hash of a file it already
+        queued: the same ``(size, mtime_ns)`` the bytes were verified with.
+        """
+        row = (
+            self.model.get_connection()
+            .execute(_QUEUED, (locator, size, mtime_ns, resolver_version))
+            .fetchone()
+        )
+        return row is not None
+
+    def review_decided_at(self, locator: str, size: int, mtime_ns: int) -> bool:
+        """True when bytes verified at this stat carry a decided item.
+
+        Any version: a newer resolver or table never adds an item beside a
+        decision, so the rescan need not copy those bytes again.
+        """
+        row = (
+            self.model.get_connection()
+            .execute(_DECIDED_AT, (locator, size, mtime_ns))
+            .fetchone()
+        )
+        return row is not None
+
+    def review_pending_at(
+        self, locator: str, size: int, mtime_ns: int, resolver_version: int
+    ) -> bool:
+        """True when a *pending* item cites bytes verified at this exact stat.
+
+        ``review_fingerprint_queued`` also matches decided items; a file whose
+        only item is decided is not in the queue.
+        """
+        row = (
+            self.model.get_connection()
+            .execute(_PENDING_AT, (locator, size, mtime_ns, resolver_version))
+            .fetchone()
+        )
+        return row is not None
+
+    def review_file_linked(self, spellings) -> bool:
+        """True when a saved row links the file under any of ``spellings``."""
+        with self._annotation_read() as conn:
+            return bool(dependent_images(conn, spellings))
+
+    def settle_linked_review_items(self, files) -> int:
+        """Remove the pending, undecided items of files that are now linked.
+
+        ``files`` holds ``(locator, spellings)`` per file. An item leaves the
+        queue only when it is still ``pending`` at revision 0 with no decision
+        row and a saved row links its file under one of ``spellings``; a
+        decided item is never touched. Files without a pending item cost one
+        indexed read and open no write transaction. Returns how many items
+        were removed (the review index revision moves when any was).
+        """
+        files = [(locator, tuple(spellings)) for locator, spellings in files]
+        for locator, _spellings in files:
+            _validate_locator(locator)
+        conn = self.model.get_connection()
+        candidates = [
+            (locator, spellings)
+            for locator, spellings in files
+            if conn.execute(_PENDING_OF_LOCATOR, (locator,)).fetchone() is not None
+        ]
+        if not candidates:
+            return 0
+        removed = 0
+        with self._pair_transaction() as conn:
+            for locator, spellings in candidates:
+                if not dependent_images(conn, spellings):
+                    continue
+                for (item_id,) in conn.execute(
+                    _PENDING_OF_LOCATOR, (locator,)
+                ).fetchall():
+                    removed += conn.execute(_SETTLE_ITEM, (item_id,)).rowcount
+            if removed:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        if removed:
+            notify_review_change(self.model.db_path)
+        return removed
+
+    def settle_stale_review_items(
+        self, locator: str, size: int, mtime_ns: int, versions
+    ) -> int:
+        """Remove pending, undecided items of the bytes verified at this stat
+        whose ``resolver_version`` is not one of ``versions``.
+
+        The caller found these bytes untraced under the current versions, so
+        an older version's item could only refuse Apply. The same safe delete
+        as the other settles; one indexed read when nothing is pending.
+        """
+        _validate_locator(locator)
+        current = set(versions)
+        conn = self.model.get_connection()
+        stale = [
+            item_id
+            for item_id, version in conn.execute(
+                _PENDING_AT_STAT, (locator, size, mtime_ns)
+            )
+            if version not in current
+        ]
+        if not stale:
+            return 0
+        removed = 0
+        with self._pair_transaction() as conn:
+            for item_id in stale:
+                removed += conn.execute(_SETTLE_ITEM, (item_id,)).rowcount
+            if removed:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        if removed:
+            notify_review_change(self.model.db_path)
+        return removed
+
+    def settle_untraced_review_items(self, locators) -> int:
+        """Remove the pending, undecided items that offer nothing to choose.
+
+        ``locators`` are files the caller found ambiguous without a choice
+        (fewer than two candidates carry a prompt). Of their items, one leaves
+        the queue only when its own stored evidence offers no choice either
+        (``evidence_offers_choice``) and it is still ``pending`` at revision 0
+        with no decision row, the same safe delete as
+        ``settle_linked_review_items``. Files without a pending item cost one
+        indexed read and open no write transaction. Returns how many items
+        were removed (the review index revision moves when any was).
+        """
+        locators = list(locators)
+        for locator in locators:
+            _validate_locator(locator)
+        conn = self.model.get_connection()
+        pending = [
+            locator
+            for locator in locators
+            if conn.execute(_PENDING_OF_LOCATOR, (locator,)).fetchone() is not None
+        ]
+        if not pending:
+            return 0
+        removed = 0
+        with self._pair_transaction() as conn:
+            for locator in pending:
+                rows = conn.execute(_PENDING_EVIDENCE_OF_LOCATOR, (locator,)).fetchall()
+                for item_id, evidence in rows:
+                    if evidence_offers_choice(json.loads(evidence)):
+                        continue
+                    removed += conn.execute(_SETTLE_ITEM, (item_id,)).rowcount
+            if removed:
+                conn.execute(_BUMP_REVIEW_INDEX)
+        return removed
+
+    def review_queue_page(self, *, limit: int, after: tuple | None = None) -> dict:
+        """One page of pending items, oldest first, from one snapshot.
+
+        ``after`` is the ``(created_at, id)`` keyset of the last item already
+        shown. Returns ``{epoch, revision, total, items, more}``; ``total`` is
+        the exact pending count and ``items`` carry their ``spelled_path`` and
+        ``locator`` (server-side only).
+        """
+        if type(limit) is not int or limit < 1:
+            raise ValueError("limit must be a positive integer")
+        with self._annotation_read() as conn:
+            epoch = conn.execute(_EPOCH).fetchone()
+            revision = conn.execute(_REVIEW_INDEX).fetchone()
+            if epoch is None or revision is None:
+                raise sqlite3.DatabaseError("Missing review queue metadata")
+            total = conn.execute(_PENDING_COUNT).fetchone()[0]
+            if after is None:
+                rows = conn.execute(_QUEUE_FIRST, (limit + 1,)).fetchall()
+            else:
+                rows = conn.execute(_QUEUE_AFTER, (*after, limit + 1)).fetchall()
+        items = [_queue_row(row, _QUEUE_COLUMNS) for row in rows[:limit]]
+        return {
+            "epoch": epoch[0],
+            "revision": revision[0],
+            "total": total,
+            "items": items,
+            "more": len(rows) > limit,
+        }
+
+    def review_item(self, item_id: str) -> dict | None:
+        """One item with its parsed evidence and the pending total, or None."""
+        if not isinstance(item_id, str) or _ITEM_ID.fullmatch(item_id) is None:
+            return None
+        with self._annotation_read() as conn:
+            epoch = conn.execute(_EPOCH).fetchone()
+            if epoch is None:
+                raise sqlite3.DatabaseError("Missing execution history metadata")
+            row = conn.execute(_DETAIL, (item_id,)).fetchone()
+            if row is None:
+                return None
+            item = _queue_row(row, _DETAIL_COLUMNS)
+            item["evidence"] = json.loads(item["evidence"])
+            item["epoch"] = epoch[0]
+            item["total"] = conn.execute(_PENDING_COUNT).fetchone()[0]
+        return item
+
+    def review_dependency_fingerprint(self, image_path: str, aliases=()) -> str:
+        with self._annotation_read() as conn:
+            return dependency_fingerprint(conn, image_path, aliases)
+
+    def review_same_name_rows(self, name: str) -> list:
+        """``(id, image_path)`` of legacy rows that may spell a file ``name``."""
+        with self._annotation_read() as conn:
+            return same_name_rows(conn, name)
+
+    def stored_prompt_titles(self, texts) -> dict:
+        """Library titles of saved prompts by text (dedup hash), when set."""
+        hashes = {generate_prompt_hash(text): text for text in texts if text.strip()}
+        if not hashes:
+            return {}
+        marks = ",".join("?" * len(hashes))
+        found: dict[str, tuple] = {}
+        conn = self.model.get_connection()
+        for prompt_hash, title in conn.execute(
+            _TITLES.format(marks=marks), list(hashes)
+        ):
+            text = hashes[prompt_hash]
+            found[text] = found.get(text, ()) + (title,)
+        return found
+
+
+def _refresh_spelling(conn: sqlite3.Connection, locator: str, spelled: str) -> bool:
+    return conn.execute(_REFRESH_SPELLING, (spelled, spelled, locator)).rowcount > 0
+
+
+def _queue_row(row, columns) -> dict:
+    item = dict(zip(columns, tuple(row)))
+    item["limited"] = bool(item["limited"])
+    return item

database/review_changes.py

@@ -0,0 +1,75 @@
+"""Needs review queue change notices (#249 PR F2).
+
+Every committed change to the pending queue (an item recorded by the scan or
+the monitor, a spelling refresh, a decision) calls ``notify_review_change``
+after its transaction, from whichever thread wrote it. The review event
+stream (``py/api/review_events.py``) subscribes, so an open Needs review page
+re-reads its queue when it changes instead of polling. Observers are keyed by
+the database file, like ``subscribe_history_restore``; an observer that fails
+is logged and never fails the write.
+"""
+
+import os
+import threading
+from typing import Callable
+
+try:
+    from ..utils.logging_config import get_logger
+except ImportError:
+    from utils.logging_config import get_logger
+
+_logger = get_logger("prompt_manager.database.review_changes")
+_lock = threading.Lock()
+_observers: dict[str, dict[object, Callable[[], None]]] = {}
+
+
+def _identity(db_path: str) -> str:
+    return os.path.normcase(os.path.realpath(db_path))
+
+
+def subscribe_review_changes(
+    db_path: str, callback: Callable[[], None]
+) -> Callable[[], None]:
+    """Call ``callback()`` after each committed queue change; return the unsubscribe."""
+    identity = _identity(db_path)
+    token = object()
+    with _lock:
+        _observers.setdefault(identity, {})[token] = callback
+
+    def unsubscribe() -> None:
+        with _lock:
+            observers = _observers.get(identity)
+            if observers is not None:
+                observers.pop(token, None)
+                if not observers:
+                    _observers.pop(identity, None)
+
+    return unsubscribe
+
+
+def observer_count(db_path: str) -> int:
+    """How many observers watch ``db_path`` (tests and diagnostics)."""
+    with _lock:
+        return len(_observers.get(_identity(db_path), {}))
+
+
+def notify_review_change(db_path: str) -> None:
+    """Tell every observer of ``db_path``; call it after the commit."""
+    with _lock:
+        callbacks = tuple(_observers.get(_identity(db_path), {}).values())
+    for callback in callbacks:
+        try:
+            callback()
+        except Exception:
+            _logger.warning("Needs review change observer failed", exc_info=True)
+
+
+def rekey_review_changes(old_path: str, new_path: str) -> None:
+    """Move the observers of ``old_path`` to ``new_path`` (Move library)."""
+    old, new = _identity(old_path), _identity(new_path)
+    if old == new:
+        return
+    with _lock:
+        observers = _observers.pop(old, None)
+        if observers:
+            _observers.setdefault(new, {}).update(observers)

database/review_decisions.py

@@ -0,0 +1,291 @@
+"""Needs review decisions (#249 E3): explicit, auditable and compatible-only.
+
+A decision is one row of ``media_review_decisions`` for one item revision.
+``selected`` links the chosen candidate's pair to the file through the public
+ingester's compatible-only rules (``_ingest_resolved_pair_in``) in the same
+``_pair_transaction`` as the decision row, the item's new state and revision
+and the review index revision; any conflict or later failure rolls all of it
+back. ``unresolved`` records the decision and links nothing. No decision
+counts a run, touches usage or calls an execution writer.
+
+Callers verify the file first (the capture must be the item's bytes and the
+current file must still hash to them); nothing here reads the filesystem.
+"""
+
+import os
+import sqlite3
+
+from .annotation_schema import BYTES_TABLE
+from .annotations import _new_id
+from .library_commands import LibraryCommandError
+from .models import utc_now_iso
+from .review import (
+    _BUMP_REVIEW_INDEX,
+    _DETAIL,
+    _DETAIL_COLUMNS,
+    _PENDING_COUNT,
+    _queue_row,
+    dependency_fingerprint,
+    dependent_images,
+    same_name_rows,
+)
+from .review_changes import notify_review_change
+from .review_schema import DECISIONS_TABLE, ITEMS_TABLE, canonical_evidence
+
+DECISION_KIND = "user-selection"
+_INSERT_DECISION = (
+    f"INSERT INTO {DECISIONS_TABLE}(id,item_id,decision_revision,"  # nosec B608
+    "created_at,outcome,candidate_hash,evidence,pair_id,image_id) "
+    "VALUES(?,?,?,?,?,?,?,?,?)"
+)
+_ADVANCE_ITEM = (
+    f"UPDATE {ITEMS_TABLE} SET state=?,revision=revision+1,"  # nosec B608
+    "updated_at=? WHERE id=? AND revision=?"
+)
+_ITEM_BYTES = (
+    f"SELECT b.sha256 FROM {ITEMS_TABLE} i "  # nosec B608
+    f"JOIN {BYTES_TABLE} b ON b.id=i.byte_id WHERE i.id=?"
+)
+
+
+class ReviewCommandError(LibraryCommandError):
+    """A decision lost a race or would rewrite an association.
+
+    ``code`` is ``review-conflict`` (the item was decided or changed since it
+    was read), ``association-conflict`` (the choice is not compatible with
+    the file's saved links, or those links changed), ``file-changed`` (the
+    item is not about these bytes) or ``review-missing``. ``item`` is the
+    current item read inside the transaction, ``total`` the pending count and
+    ``fingerprint`` the current dependency fingerprint, when known;
+    ``changed`` marks an association conflict caused by links that changed
+    since the reviewer read them (rather than an incompatible choice).
+    """
+
+    def __init__(
+        self, code, epoch, *, item=None, total=None, fingerprint=None, changed=False
+    ):
+        super().__init__(code, epoch)
+        self.item = item
+        self.total = total
+        self.fingerprint = fingerprint
+        self.changed = changed
+        # The public view of ``item``, set by the API when it formats one.
+        self.view = None
+
+
+class _Rollback(Exception):
+    """Raised inside the transaction so the context manager rolls back."""
+
+    def __init__(self, error: ReviewCommandError):
+        super().__init__(error.code)
+        self.error = error
+
+
+class ReviewDecisionOperations:
+    """PromptDatabase mixin: record one decision on one Needs review item."""
+
+    def decide_review_item(
+        self,
+        *,
+        item_id: str,
+        expected_epoch: str,
+        expected_revision: int,
+        sha256: str,
+        outcome: str,
+        candidate: dict | None = None,
+        image_path: str | None = None,
+        metadata: dict | None = None,
+        expected_fingerprint: str | None = None,
+        aliases: tuple = (),
+        known_rows: frozenset | None = None,
+    ) -> dict:
+        """Record the decision; return ``{item, total, pair}``.
+
+        ``selected`` needs the evidence ``candidate`` (with its ``hash``), the
+        ``image_path`` the scan and the monitor link under, the captured
+        ``metadata`` and the ``expected_fingerprint`` the reviewer saw.
+        ``aliases`` are the other spellings legacy rows of the same physical
+        file use (a symlinked, junctioned or mapped root) and ``known_rows``
+        the ids of every same-name row the caller classified; a same-name row
+        it did not see is a changed association.
+        Raises ``LibraryCommandError("storage-changed")`` after a restore and
+        :class:`ReviewCommandError` for every refused decision, with zero
+        writes either way.
+        """
+        if outcome not in ("selected", "unresolved"):
+            raise ValueError("outcome must be selected or unresolved")
+        if outcome == "selected" and (
+            not isinstance(candidate, dict) or not image_path
+        ):
+            raise ValueError("A selected decision needs a candidate and a file")
+        try:
+            with self._pair_transaction() as conn:
+                decided = self._decide_in(
+                    conn,
+                    item_id=item_id,
+                    expected_epoch=expected_epoch,
+                    expected_revision=expected_revision,
+                    sha256=sha256,
+                    outcome=outcome,
+                    candidate=candidate,
+                    image_path=image_path,
+                    metadata=metadata or {},
+                    expected_fingerprint=expected_fingerprint,
+                    aliases=tuple(aliases),
+                    known_rows=known_rows,
+                )
+        except _Rollback as rollback:
+            raise rollback.error from None
+        # After the commit: open Needs review pages re-read their queue.
+        notify_review_change(self.model.db_path)
+        return decided
+
+    def _decide_in(self, conn: sqlite3.Connection, **command) -> dict:
+        epoch = self._library_epoch(conn, command["expected_epoch"])
+        item_id = command["item_id"]
+        current = _current_item(conn, item_id)
+        if current is None:
+            raise _Rollback(ReviewCommandError("review-missing", epoch))
+        total = conn.execute(_PENDING_COUNT).fetchone()[0]
+        if (
+            current["state"] != "pending"
+            or current["revision"] != command["expected_revision"]
+        ):
+            raise _Rollback(
+                ReviewCommandError("review-conflict", epoch, item=current, total=total)
+            )
+        if conn.execute(_ITEM_BYTES, (item_id,)).fetchone()[0] != command["sha256"]:
+            raise _Rollback(ReviewCommandError("file-changed", epoch))
+        decision_id = _new_id()
+        pair = None
+        candidate = command["candidate"] if command["outcome"] == "selected" else None
+        if candidate is not None:
+            pair = self._apply_candidate(
+                conn, epoch, current, decision_id, total, **command
+            )
+        evidence = {
+            "kind": DECISION_KIND,
+            "decision_id": decision_id,
+            "outcome": command["outcome"],
+            "reviewed_digest": command["sha256"],
+            "item_evidence_hash": current["evidence_hash"],
+            "resolver_version": current["resolver_version"],
+            "candidate": candidate,
+            "dependency_fingerprint": command["expected_fingerprint"],
+            "pair": pair,
+        }
+        now = utc_now_iso()
+        self._insert_review_decision(
+            conn,
+            (
+                decision_id,
+                item_id,
+                current["revision"] + 1,
+                now,
+                command["outcome"],
+                candidate["hash"] if candidate is not None else None,
+                canonical_evidence(evidence),
+                pair["id"] if pair else None,
+                pair["image_id"] if pair else None,
+            ),
+        )
+        conn.execute(
+            _ADVANCE_ITEM, (command["outcome"], now, item_id, current["revision"])
+        )
+        conn.execute(_BUMP_REVIEW_INDEX)
+        return {
+            "item": _current_item(conn, item_id),
+            "total": conn.execute(_PENDING_COUNT).fetchone()[0],
+            "pair": pair,
+            "decision_id": decision_id,
+        }
+
+    def _apply_candidate(
+        self, conn, epoch, current, decision_id, total, **command
+    ) -> dict:
+        """Link the candidate's pair, compatible-only; raise on any conflict."""
+        # Imported here: pair_ingestion imports operations, which composes this.
+        from .pair_ingestion import _ingest_resolved_pair_in
+
+        image_path = command["image_path"]
+        aliases = command["aliases"]
+        fingerprint = dependency_fingerprint(conn, image_path, aliases)
+        known = command["known_rows"]
+        unseen = known is not None and any(
+            row_id not in known
+            for row_id, _path in same_name_rows(conn, os.path.basename(image_path))
+        )
+        if unseen or fingerprint != command["expected_fingerprint"]:
+            raise _Rollback(
+                ReviewCommandError(
+                    "association-conflict",
+                    epoch,
+                    item=current,
+                    total=total,
+                    fingerprint=fingerprint,
+                    changed=True,
+                )
+            )
+        rows = dependent_images(conn, (image_path, *aliases))
+        if len(rows) > 1:
+            # One file linked under several spellings: no single row to enrich.
+            raise _Rollback(
+                ReviewCommandError(
+                    "association-conflict",
+                    epoch,
+                    item=current,
+                    total=total,
+                    fingerprint=fingerprint,
+                )
+            )
+        if rows:
+            # Enrich the one row of this file under the spelling it was
+            # linked with, so the ingester's compatibility rules see it.
+            image_path = rows[0][1]
+        candidate = command["candidate"]
+        positive, negative = candidate["positive"], candidate["negative"]
+        provenance = {
+            "kind": DECISION_KIND,
+            "decision_id": decision_id,
+            "reviewed_digest": command["sha256"],
+            "candidate_hash": candidate["hash"],
+            "resolver_version": current["resolver_version"],
+            "positive": positive,
+            "negative": negative,
+        }
+        result = _ingest_resolved_pair_in(
+            conn,
+            self,
+            image_path,
+            command["metadata"],
+            {"state": "resolved", "positive": positive, "negative": negative},
+            provenance=provenance,
+        )
+        if result["status"] == "conflict":
+            raise _Rollback(
+                ReviewCommandError(
+                    "association-conflict",
+                    epoch,
+                    item=current,
+                    total=total,
+                    fingerprint=fingerprint,
+                )
+            )
+        return {
+            "id": result["pair_id"],
+            "image_id": result["image_id"],
+            "status": result["status"],
+        }
+
+    @staticmethod
+    def _insert_review_decision(conn: sqlite3.Connection, values: tuple) -> None:
+        conn.execute(_INSERT_DECISION, values)
+
+
+def _current_item(conn: sqlite3.Connection, item_id: str) -> dict | None:
+    row = conn.execute(_DETAIL, (item_id,)).fetchone()
+    if row is None:
+        return None
+    item = _queue_row(row, _DETAIL_COLUMNS)
+    item.pop("evidence")
+    return item

database/review_schema.py

@@ -0,0 +1,513 @@
+"""Needs review schema group (#249): capture-bound ambiguous items and decisions.
+
+The group installs like ``image_annotations``: optional in old backups,
+required once a database is current, and valid only beside a valid image
+annotation group, because every item cites one annotation byte record. It owns
+
+- ``media_review_metadata``: one row, the group's schema version and the
+  review index revision (moved by every queue change);
+- ``media_review_items``: one ambiguous resolver result for one byte record,
+  keyed by ``(byte_id, resolver_version, evidence_hash)``, with the file's
+  spelling when it was recorded (the locator is case-folded on Windows, a
+  target is not), its canonical evidence, candidate count, ``limited`` flag,
+  state and revision;
+- ``media_review_decisions``: the auditable decision history of an item, one
+  row per decision revision, with optional navigation to the pair and image
+  the decision linked (nulled, never cascaded, when those are deleted).
+
+Evidence is canonical JSON (:func:`canonical_evidence`): one object, sorted
+keys, no whitespace, ASCII only (so characters, bytes and the 8 MiB bound
+agree and lone surrogates survive), no NaN or infinity. ``evidence_hash`` is
+the SHA-256 of the item's evidence text.
+
+Installation never backfills anything and never touches the filesystem.
+Validation reads only: catalog objects compared after whitespace compaction,
+columns and foreign keys through the PRAGMAs, persisted values in single
+bounded SQL passes, then one streamed pass that re-canonicalizes each evidence
+document (one row in memory at a time).
+"""
+
+import hashlib
+import json
+import re
+import sqlite3
+from typing import Any, Optional
+
+from .annotation_schema import BYTES_TABLE
+from .annotation_schema import TABLES as ANNOTATION_TABLES
+from .annotation_schema import _catalog as _annotation_catalog
+from .annotation_schema import annotation_schema_error
+
+try:
+    from ..utils.json_strict import dumps_strict
+except ImportError:
+    from utils.json_strict import dumps_strict
+
+SCHEMA_VERSION = 1
+METADATA_TABLE = "media_review_metadata"
+ITEMS_TABLE = "media_review_items"
+DECISIONS_TABLE = "media_review_decisions"
+TABLES = (METADATA_TABLE, ITEMS_TABLE, DECISIONS_TABLE)
+MAX_EVIDENCE_BYTES = 8 * 1024 * 1024
+MAX_CANDIDATES = 16
+# The file's own spelling when it was recorded (bytes of UTF-8 text). Locators
+# are case-folded on Windows, so a target is formatted from this instead.
+MAX_SPELLED_PATH_BYTES = 16384
+ITEM_STATES = ("pending", "selected", "unresolved")
+DECISION_OUTCOMES = ("selected", "unresolved")
+_TIMESTAMP_BYTES = 64
+_HEX64 = "length({c})=64 AND {c} NOT GLOB '*[^0-9a-f]*'"
+_EVIDENCE_CHECK = f"CHECK(length(CAST(evidence AS BLOB)) <= {MAX_EVIDENCE_BYTES})"
+
+_TABLE_DDL = {
+    METADATA_TABLE: (
+        f"CREATE TABLE {METADATA_TABLE} (singleton_id INTEGER PRIMARY KEY "
+        "CHECK(singleton_id=1), schema_version INTEGER NOT NULL "
+        f"CHECK(schema_version={SCHEMA_VERSION}), "
+        "review_index_revision INTEGER NOT NULL DEFAULT 0 "
+        "CHECK(review_index_revision>=0))"
+    ),
+    ITEMS_TABLE: (
+        f"CREATE TABLE {ITEMS_TABLE} (id TEXT PRIMARY KEY CHECK(length(id)=36), "
+        f"byte_id TEXT NOT NULL REFERENCES {BYTES_TABLE}(id) ON DELETE RESTRICT, "
+        "spelled_path TEXT NOT NULL CHECK(length(CAST(spelled_path AS BLOB)) "
+        f"BETWEEN 1 AND {MAX_SPELLED_PATH_BYTES}), "
+        "resolver_version INTEGER NOT NULL, evidence_hash TEXT NOT NULL "
+        f"CHECK({_HEX64.format(c='evidence_hash')}), "
+        f"evidence TEXT NOT NULL {_EVIDENCE_CHECK}, "
+        "candidate_count INTEGER NOT NULL "
+        f"CHECK(candidate_count BETWEEN 0 AND {MAX_CANDIDATES}), "
+        "limited INTEGER NOT NULL CHECK(limited IN (0,1)), "
+        "state TEXT NOT NULL CHECK(state IN ('pending','selected','unresolved')), "
+        "revision INTEGER NOT NULL CHECK(revision>=0), created_at TEXT NOT NULL, "
+        "updated_at TEXT NOT NULL, UNIQUE(byte_id, resolver_version, evidence_hash))"
+    ),
+    DECISIONS_TABLE: (
+        f"CREATE TABLE {DECISIONS_TABLE} (id TEXT PRIMARY KEY "
+        "CHECK(length(id)=36), item_id TEXT NOT NULL "
+        f"REFERENCES {ITEMS_TABLE}(id) ON DELETE RESTRICT, "
+        "decision_revision INTEGER NOT NULL, created_at TEXT NOT NULL, "
+        "outcome TEXT NOT NULL CHECK(outcome IN ('selected','unresolved')), "
+        "candidate_hash TEXT CHECK(candidate_hash IS NULL OR "
+        f"({_HEX64.format(c='candidate_hash')})), "
+        f"evidence TEXT NOT NULL {_EVIDENCE_CHECK}, "
+        "pair_id INTEGER REFERENCES prompt_pairs(id) ON DELETE SET NULL, "
+        "image_id INTEGER REFERENCES generated_images(id) ON DELETE SET NULL, "
+        "UNIQUE(item_id, decision_revision), "
+        "CHECK((outcome='selected') = (candidate_hash IS NOT NULL)))"
+    ),
+}
+_INDEX_DDL = {
+    # The queue reads pending items oldest first with a stable tiebreak.
+    "idx_media_review_items_queue": (
+        "CREATE INDEX idx_media_review_items_queue ON "
+        f"{ITEMS_TABLE}(state, created_at, id)"
+    ),
+    # Deleting a pair or an image nulls decision navigation by index, not scan.
+    "idx_media_review_decisions_pair": (
+        "CREATE INDEX idx_media_review_decisions_pair ON " f"{DECISIONS_TABLE}(pair_id)"
+    ),
+    "idx_media_review_decisions_image": (
+        "CREATE INDEX idx_media_review_decisions_image ON "
+        f"{DECISIONS_TABLE}(image_id)"
+    ),
+}
+# ``sqlite_autoindex_<table>_<n>``: the TEXT primary key and the UNIQUE key.
+_AUTOINDEXES = {ITEMS_TABLE: 2, DECISIONS_TABLE: 2}
+# (name, declared type, not null, default, primary key position)
+_COLUMNS = {
+    METADATA_TABLE: (
+        ("singleton_id", "INTEGER", 0, None, 1),
+        ("schema_version", "INTEGER", 1, None, 0),
+        ("review_index_revision", "INTEGER", 1, "0", 0),
+    ),
+    ITEMS_TABLE: (
+        ("id", "TEXT", 0, None, 1),
+        ("byte_id", "TEXT", 1, None, 0),
+        ("spelled_path", "TEXT", 1, None, 0),
+        ("resolver_version", "INTEGER", 1, None, 0),
+        ("evidence_hash", "TEXT", 1, None, 0),
+        ("evidence", "TEXT", 1, None, 0),
+        ("candidate_count", "INTEGER", 1, None, 0),
+        ("limited", "INTEGER", 1, None, 0),
+        ("state", "TEXT", 1, None, 0),
+        ("revision", "INTEGER", 1, None, 0),
+        ("created_at", "TEXT", 1, None, 0),
+        ("updated_at", "TEXT", 1, None, 0),
+    ),
+    DECISIONS_TABLE: (
+        ("id", "TEXT", 0, None, 1),
+        ("item_id", "TEXT", 1, None, 0),
+        ("decision_revision", "INTEGER", 1, None, 0),
+        ("created_at", "TEXT", 1, None, 0),
+        ("outcome", "TEXT", 1, None, 0),
+        ("candidate_hash", "TEXT", 0, None, 0),
+        ("evidence", "TEXT", 1, None, 0),
+        ("pair_id", "INTEGER", 0, None, 0),
+        ("image_id", "INTEGER", 0, None, 0),
+    ),
+}
+# (column, parent table, parent column, on delete)
+_FOREIGN_KEYS = {
+    METADATA_TABLE: set(),
+    ITEMS_TABLE: {("byte_id", BYTES_TABLE, "id", "RESTRICT")},
+    DECISIONS_TABLE: {
+        ("item_id", ITEMS_TABLE, "id", "RESTRICT"),
+        ("pair_id", "prompt_pairs", "id", "SET NULL"),
+        ("image_id", "generated_images", "id", "SET NULL"),
+    },
+}
+
+
+class _NonCanonical(ValueError):
+    pass
+
+
+def _reject_constant(name: str) -> Any:
+    raise _NonCanonical(f"JSON constant {name} is not allowed")
+
+
+# Built once: validation re-encodes every stored document at startup.
+_ENCODER = json.JSONEncoder(
+    sort_keys=True, separators=(",", ":"), ensure_ascii=True, allow_nan=False
+)
+_DECODER = json.JSONDecoder(parse_constant=_reject_constant)
+
+
+def canonical_evidence(value: Any) -> str:
+    """Serialize one evidence object canonically; ``ValueError`` when invalid.
+
+    The text is a JSON object with sorted keys, no whitespace, ASCII only and
+    no NaN/infinity, at most ``MAX_EVIDENCE_BYTES`` bytes. A non-finite number
+    (ComfyUI stamps ``NaN`` into graphs) is written as ``null``, the store's
+    policy everywhere (``utils.json_strict.dumps_strict``).
+    """
+    if not isinstance(value, dict):
+        raise ValueError("Review evidence must be a JSON object")
+    try:
+        try:
+            text = _ENCODER.encode(value)
+        except ValueError as error:
+            if "Out of range float" not in str(error):
+                raise
+            text = dumps_strict(
+                value, sort_keys=True, separators=(",", ":"), ensure_ascii=True
+            )
+    except (TypeError, RecursionError) as error:
+        raise ValueError(f"Review evidence is not serializable: {error}") from None
+    if len(text) > MAX_EVIDENCE_BYTES:
+        raise ValueError("Review evidence is larger than 8 MiB")
+    return text
+
+
+def choosable(candidate: Any) -> bool:
+    """True when a candidate offers a prompt to choose: it is ``resolved`` and
+    its positive side carries non-blank text. The page disables any other
+    candidate, so it is never something to choose.
+    """
+    if not isinstance(candidate, dict) or candidate.get("state") != "resolved":
+        return False
+    positive = candidate.get("positive")
+    if not isinstance(positive, dict):
+        return False
+    text = positive.get("text")
+    return isinstance(text, str) and bool(text.strip())
+
+
+# An item is worth a decision only when at least this many candidates offer
+# a prompt; with fewer there is nothing to choose between (maintainer test of
+# #284/#288: 39 items whose candidates were all "unresolved").
+MIN_CHOOSABLE = 2
+
+
+def evidence_offers_choice(evidence: Any) -> bool:
+    """False when stored evidence provably offers nothing to choose.
+
+    Unlimited evidence must list ``MIN_CHOOSABLE`` choosable candidates. A
+    limited item stores none, only ``candidate_total``; it may offer a choice
+    (more than ``MAX_CANDIDATES`` prompts) unless that total is too small.
+    """
+    if not isinstance(evidence, dict):
+        return False
+    if evidence.get("limited"):
+        total = evidence.get("candidate_total")
+        return type(total) is int and total >= MIN_CHOOSABLE
+    candidates = evidence.get("candidates")
+    if not isinstance(candidates, list):
+        return False
+    return sum(1 for c in candidates if choosable(c)) >= MIN_CHOOSABLE
+
+
+def evidence_digest(text: str) -> str:
+    """SHA-256 of canonical evidence text (ASCII, hence UTF-8 identical)."""
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def _is_canonical(text: Any) -> bool:
+    if not isinstance(text, str):
+        return False
+    try:
+        value = _DECODER.decode(text)
+        return canonical_evidence(value) == text
+    except (ValueError, RecursionError):
+        return False
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql).lower()
+
+
+def _uuid(column: str) -> str:
+    """SQL: *column* is a lowercase hyphenated 36-character UUID string."""
+    return (
+        f"typeof({column}) = 'text' AND length({column}) = 36 "
+        f"AND substr({column}, 9, 1) = '-' AND substr({column}, 14, 1) = '-' "
+        f"AND substr({column}, 19, 1) = '-' AND substr({column}, 24, 1) = '-' "
+        f"AND length(replace({column}, '-', '')) = 32 "
+        f"AND replace({column}, '-', '') NOT GLOB '*[^0-9a-f]*'"
+    )
+
+
+def _timestamp(column: str) -> str:
+    return (
+        f"typeof({column}) = 'text' "
+        f"AND length(CAST({column} AS BLOB)) <= {_TIMESTAMP_BYTES}"
+    )
+
+
+def _non_negative(column: str) -> str:
+    return f"typeof({column}) = 'integer' AND {column} >= 0"
+
+
+def _hex64(column: str) -> str:
+    return f"typeof({column}) = 'text' AND " + _HEX64.format(c=column)
+
+
+def _evidence(column: str) -> str:
+    # Type and size only: the streamed pass below proves the text canonical
+    # (which implies printable ASCII), so a pattern scan here would read every
+    # document a second time.
+    return (
+        f"typeof({column}) = 'text' "
+        f"AND length(CAST({column} AS BLOB)) BETWEEN 2 AND {MAX_EVIDENCE_BYTES}"
+    )
+
+
+def _nullable_integer(column: str) -> str:
+    return f"({column} IS NULL OR typeof({column}) = 'integer')"
+
+
+def _in(column: str, values: tuple) -> str:
+    return f"typeof({column}) = 'text' AND {column} IN ({repr(values)[1:-1]})"
+
+
+# Every row of a table must satisfy its predicate; one pass each, LIMIT 1.
+_ROW_PREDICATES = {
+    ITEMS_TABLE: " AND ".join(
+        (
+            _uuid("id"),
+            _uuid("byte_id"),
+            "typeof(spelled_path) = 'text' AND length(CAST(spelled_path AS BLOB)) "
+            f"BETWEEN 1 AND {MAX_SPELLED_PATH_BYTES}",
+            "typeof(resolver_version) = 'integer'",
+            _hex64("evidence_hash"),
+            _evidence("evidence"),
+            _non_negative("candidate_count"),
+            f"candidate_count <= {MAX_CANDIDATES}",
+            "typeof(limited) = 'integer' AND limited IN (0, 1)",
+            _in("state", ITEM_STATES),
+            _non_negative("revision"),
+            _timestamp("created_at"),
+            _timestamp("updated_at"),
+        )
+    ),
+    DECISIONS_TABLE: " AND ".join(
+        (
+            _uuid("id"),
+            _uuid("item_id"),
+            _non_negative("decision_revision"),
+            _timestamp("created_at"),
+            _in("outcome", DECISION_OUTCOMES),
+            "(CASE WHEN outcome = 'selected' THEN "
+            f"{_hex64('candidate_hash')} ELSE candidate_hash IS NULL END)",
+            _evidence("evidence"),
+            _nullable_integer("pair_id"),
+            _nullable_integer("image_id"),
+        )
+    ),
+}
+
+
+def _expected_catalog() -> set:
+    objects = {("table", name, _compact(sql)) for name, sql in _TABLE_DDL.items()}
+    objects |= {("index", name, _compact(sql)) for name, sql in _INDEX_DDL.items()}
+    objects |= {
+        ("index", f"sqlite_autoindex_{table}_{n}", "")
+        for table, count in _AUTOINDEXES.items()
+        for n in range(1, count + 1)
+    }
+    return objects
+
+
+def _catalog(conn: sqlite3.Connection) -> list:
+    names = TABLES + tuple(_INDEX_DDL)
+    marks = ",".join("?" * len(names))
+    table_marks = ",".join("?" * len(TABLES))
+    return conn.execute(  # only "?" placeholders are interpolated
+        "SELECT type,name,sql FROM sqlite_master "
+        f"WHERE name IN ({marks}) OR tbl_name IN ({table_marks})",  # nosec B608
+        names + TABLES,
+    ).fetchall()
+
+
+def _structure_error(conn: sqlite3.Connection) -> Optional[str]:
+    for table, expected in _COLUMNS.items():
+        columns = tuple(
+            (row[1], row[2].upper(), row[3], row[4], row[5])
+            for row in conn.execute(f"PRAGMA table_info({table})")
+        )
+        if columns != expected:
+            return f"Review queue table {table} has invalid columns"
+        foreign_keys = {
+            (row[3], row[2], row[4], row[6])
+            for row in conn.execute(f"PRAGMA foreign_key_list({table})")
+        }
+        if foreign_keys != _FOREIGN_KEYS[table]:
+            return f"Review queue table {table} has invalid foreign keys"
+        if conn.execute(f"PRAGMA foreign_key_check({table})").fetchone():
+            return f"Review queue table {table} contains foreign key violations"
+    return None
+
+
+def _evidence_error(conn: sqlite3.Connection) -> Optional[str]:
+    """Stream every evidence document once; one row is held at a time.
+
+    Rows are read with ``fetchone``: iterating a managed connection's cursor
+    would prefetch ``ITER_CHUNK`` (128) rows (``database/managed_connection.py``).
+    """
+    items = conn.execute(
+        f"SELECT evidence_hash, evidence FROM {ITEMS_TABLE}"  # nosec B608
+    )
+    while (row := items.fetchone()) is not None:
+        evidence_hash, text = row[0], row[1]
+        if not _is_canonical(text) or evidence_digest(text) != evidence_hash:
+            return f"Review queue table {ITEMS_TABLE} contains invalid evidence"
+    decisions = conn.execute(f"SELECT evidence FROM {DECISIONS_TABLE}")  # nosec B608
+    while (row := decisions.fetchone()) is not None:
+        text = row[0]
+        if not _is_canonical(text):
+            return f"Review queue table {DECISIONS_TABLE} contains invalid evidence"
+    return None
+
+
+def _value_error(conn: sqlite3.Connection) -> Optional[str]:
+    # Only this module's constants are interpolated into the SQL below.
+    rows = conn.execute(
+        "SELECT COUNT(*), MIN(typeof(singleton_id) = 'integer' "  # nosec B608
+        "AND singleton_id = 1 AND typeof(schema_version) = 'integer' "
+        f"AND schema_version = {SCHEMA_VERSION} "
+        f"AND {_non_negative('review_index_revision')}) FROM {METADATA_TABLE}"
+    ).fetchone()
+    if rows[0] != 1 or rows[1] != 1:
+        return "Review queue metadata is invalid"
+    for table, predicate in _ROW_PREDICATES.items():
+        if conn.execute(
+            f"SELECT 1 FROM {table} WHERE NOT ({predicate}) LIMIT 1"  # nosec B608
+        ).fetchone():
+            return f"Review queue table {table} contains invalid values"
+    return _evidence_error(conn)
+
+
+def review_without_annotations_error(conn: sqlite3.Connection) -> Optional[str]:
+    """Catalog only: a review group present without its annotation tables.
+
+    Startup asks this before the annotation group installs, which would make
+    a review group copied without it look valid. It reads no rows; the full
+    validation runs once in :func:`migrate_review_state`.
+    """
+    if not _catalog(conn):
+        return None
+    present = {
+        name for kind, name, _sql in _annotation_catalog(conn) if kind == "table"
+    }
+    if present != set(ANNOTATION_TABLES):
+        return (
+            "Review queue schema requires a valid image annotation schema: "
+            "Missing image annotation schema"
+        )
+    return None
+
+
+def review_schema_error(
+    conn: sqlite3.Connection,
+    *,
+    allow_missing: bool = True,
+    annotations_valid: bool = False,
+) -> Optional[str]:
+    """Reject partial or unsupported review schemas and invalid rows.
+
+    Reads only: nothing is ever repaired. A database without any of the
+    group's objects is acceptable for backups that predate it unless
+    ``allow_missing`` is false. A present group requires a present, valid
+    image annotation group; ``annotations_valid`` says the caller has just
+    validated it (startup's annotation install), so only its presence is
+    checked again.
+    """
+    objects = _catalog(conn)
+    if not objects:
+        return None if allow_missing else "Missing review queue schema"
+    if annotations_valid:
+        annotation_error = review_without_annotations_error(conn)
+    else:
+        annotation_error = annotation_schema_error(conn, allow_missing=False)
+        if annotation_error:
+            annotation_error = (
+                "Review queue schema requires a valid image annotation schema: "
+                + annotation_error
+            )
+    if annotation_error:
+        return annotation_error
+    present = {name for kind, name, _sql in objects if kind == "table"}
+    if present != set(TABLES):
+        return "Incomplete review queue schema"
+    actual = {(kind, name, _compact(sql or "")) for kind, name, sql in objects}
+    if actual != _expected_catalog():
+        return "Review queue has an invalid schema definition"
+    return _structure_error(conn) or _value_error(conn)
+
+
+def migrate_review_state(conn: sqlite3.Connection) -> None:
+    """Install the group atomically; validate before and after creating it.
+
+    Runs after ``migrate_annotation_state`` (which has just validated the
+    annotation group): the group is only valid beside the annotation group it
+    cites. A present group is validated once, evidence included.
+    """
+    conn.execute("SAVEPOINT media_review_schema")
+    try:
+        installed = bool(_catalog(conn))
+        error = review_schema_error(conn, annotations_valid=True)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        if installed:
+            # Already complete and valid: nothing to create, and the evidence
+            # pass is the expensive part of validation, so it runs once.
+            conn.execute("RELEASE media_review_schema")
+            return
+        for sql in _TABLE_DDL.values():
+            conn.execute(sql.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        for sql in _INDEX_DDL.values():
+            conn.execute(sql.replace("INDEX", "INDEX IF NOT EXISTS", 1))
+        conn.execute(
+            f"INSERT OR IGNORE INTO {METADATA_TABLE} "
+            "(singleton_id, schema_version, review_index_revision) VALUES (1, ?, 0)",
+            (SCHEMA_VERSION,),
+        )
+        error = review_schema_error(conn, allow_missing=False, annotations_valid=True)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute("RELEASE media_review_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO media_review_schema")
+        conn.execute("RELEASE media_review_schema")
+        raise

database/safety_backup.py

@@ -0,0 +1,185 @@
+"""Safety backups beside the live database: one writer for every caller.
+
+A safety backup is ``<db>.<label>_<YYYYmmdd_HHMMSS_ffffff>[_n]`` in the
+library's own folder. It is written by SQLite's backup API into a temp file
+beside the database (``SAFETY_TEMP_PREFIX``, swept at startup after a crash),
+made one self-contained file (rollback journal), optionally verified, fsynced,
+made private (``0o600`` on POSIX) and only then given its unique name, so a
+backup name never carries a half-written file. The names are unique within
+one second (microseconds, then a counter): on Windows two backups in the same
+second must not overwrite each other.
+
+Maintenance applies take a verified backup (``py/api/maintenance.py``); a
+restore takes an unverified one, because the database it replaces may be the
+damaged one the user is recovering from (``database/restore_stage.py``).
+"""
+
+import datetime
+import os
+import sqlite3
+import tempfile
+from contextlib import closing
+
+SAFETY_TEMP_PREFIX = ".pm-safety-"
+_NAME_ATTEMPTS = 1000
+
+
+class SafetyBackupFailed(Exception):
+    """The safety backup could not be written, verified or put in place.
+
+    ``reason`` is the verify reason when the copy did not verify.
+    """
+
+    def __init__(self, message, reason=None):
+        super().__init__(message)
+        self.reason = reason
+
+
+def _now():
+    return datetime.datetime.now()
+
+
+def remove_quietly(*paths):
+    """Remove each existing path, ignoring errors (cleanup after a failure)."""
+    for path in paths:
+        try:
+            if path and os.path.exists(path):
+                os.remove(path)
+        except OSError:
+            pass
+
+
+def backup_names(db_path, label="backup"):
+    """Candidate names, unique within one second (microseconds, then ``_n``)."""
+    base = f"{db_path}.{label}_{_now().strftime('%Y%m%d_%H%M%S_%f')}"
+    for attempt in range(_NAME_ATTEMPTS):
+        yield base if attempt == 0 else f"{base}_{attempt}"
+
+
+def reserve_backup_name(db_path, label="backup"):
+    """Create and return an unused ``<db>.<label>_<stamp>[_n]`` (O_EXCL)."""
+    for candidate in backup_names(db_path, label):
+        try:
+            fd = os.open(candidate, os.O_CREAT | os.O_EXCL | os.O_WRONLY, 0o600)
+        except FileExistsError:
+            continue
+        os.close(fd)
+        return candidate
+    raise SafetyBackupFailed("No free backup name")
+
+
+def place_backup(temp, db_path, label="backup"):
+    """Give the finished ``temp`` its unique backup name; return that path.
+
+    ``os.link`` is atomic and refuses an existing name, so no empty or half
+    file ever carries a backup name, even across a crash. A filesystem
+    without hard links falls back to reserving the name (``O_EXCL``) and
+    ``os.replace``. The temp file is removed by the caller.
+    """
+    for candidate in backup_names(db_path, label):
+        try:
+            os.link(temp, candidate)
+        except FileExistsError:
+            continue
+        except (OSError, NotImplementedError):
+            break  # no hard links here
+        else:
+            return candidate
+    else:
+        raise SafetyBackupFailed("No free backup name")
+    final = reserve_backup_name(db_path, label)
+    try:
+        os.replace(temp, final)
+    except OSError:
+        remove_quietly(final)
+        raise
+    return final
+
+
+def sweep_safety_temps(db_path):
+    """Remove the temp copies an interrupted safety backup left (blocking).
+
+    Called once at startup, before any job can take a backup. Returns the
+    names removed.
+    """
+    folder = os.path.dirname(os.path.abspath(db_path))
+    removed = []
+    try:
+        names = os.listdir(folder)
+    except OSError:
+        return removed
+    for name in names:
+        if not (name.startswith(SAFETY_TEMP_PREFIX) and ".tmp" in name):
+            continue
+        try:
+            os.remove(os.path.join(folder, name))
+            removed.append(name)
+        except OSError:
+            continue
+    return removed
+
+
+def make_standalone(path):
+    """Make the copy one self-contained file (rollback journal, not WAL).
+
+    The backup API copies the live database's WAL flag; a WAL file opened
+    read-only later (a verify, a restore preview) would leave ``-wal`` and
+    ``-shm`` sidecars beside it.
+    """
+    try:
+        with closing(sqlite3.connect(path)) as conn:
+            conn.execute("PRAGMA journal_mode=DELETE").fetchone()
+    except sqlite3.Error as exc:
+        raise SafetyBackupFailed("The backup could not be finished") from exc
+
+
+def fsync_file(path):
+    with open(path, "rb+") as handle:
+        os.fsync(handle.fileno())
+
+
+def safety_temp(db_path):
+    """A new, closed temp file beside the database (``SAFETY_TEMP_PREFIX``)."""
+    fd, temp = tempfile.mkstemp(
+        dir=os.path.dirname(os.path.abspath(db_path)),
+        prefix=SAFETY_TEMP_PREFIX,
+        suffix=".tmp",
+    )
+    os.close(fd)
+    return temp
+
+
+def take_safety_backup(model, label="backup", *, verify=True):
+    """Write (and by default verify) a safety backup; return its path (blocking).
+
+    The copy is made in a temp file beside the database and only renamed
+    onto its final, unique name ``<db>.<label>_<stamp>`` once it is complete
+    (and verified), so a backup file on disk is always a whole one. ``label``
+    "backup" joins the pruned pool; another label (the workflow storage
+    migration's ``pre-workflow-storage``) is kept until the user removes it.
+    Raises SafetyBackupFailed, leaving no file.
+    """
+    db_path = os.path.abspath(model.db_path)
+    temp = safety_temp(db_path)
+    final = None
+    try:
+        if not model.backup_database(temp):
+            raise SafetyBackupFailed("The backup could not be written")
+        make_standalone(temp)
+        if verify:
+            ok, reason = model.verify_database_file(temp)
+            if not ok:
+                raise SafetyBackupFailed("The backup did not verify", reason)
+        fsync_file(temp)
+        if os.name != "nt":
+            os.chmod(temp, 0o600)
+        final = place_backup(temp, db_path, label)
+        return final
+    except OSError as exc:
+        remove_quietly(final)
+        raise SafetyBackupFailed("The backup could not be put in place") from exc
+    except SafetyBackupFailed:
+        remove_quietly(final)
+        raise
+    finally:
+        remove_quietly(temp, temp + "-journal", temp + "-wal", temp + "-shm")

database/saved_views.py

@@ -0,0 +1,334 @@
+"""Persisted saved views (D16): a named Images or Library toolbar state.
+
+A view's state holds exactly its page's toolbar values and its Refine-this-view
+refinements, whitelisted per page and stored as compact JSON of at most 4 KB.
+
+Writes (the API's POST and PATCH) validate strictly. Startup and restore check
+only the structure (the table, its columns and each row's types, sizes and JSON
+object), never the value rules: a row saved under older rules must not stop the
+database from opening. Startup is more tolerant still: a row whose state is not
+a JSON object is logged by id, left in place and skipped by every read, while
+restore rejects such a backup. Reads normalize instead, so a stale row opens with its
+usable values and defaults for the rest. Optional in old backups, required once
+a database is current (see ``library_schema.py``).
+"""
+
+import json
+import logging
+import re
+import sqlite3
+from typing import Optional
+
+# The models logger's parent ("prompt_manager.database.models" logs there too).
+_LOGGER = logging.getLogger("prompt_manager.database.saved_views")
+
+TABLE = "saved_views"
+PAGES = ("images", "library")
+# The schema check's literal SQL repeats the name and state caps (names in
+# bytes, at most 4 per character); keep them in step.
+MAX_NAME_LENGTH = 100
+MAX_STATE_BYTES = 4096
+MAX_TAGS = 20
+MAX_TAG_LENGTH = 50
+_IMAGE_SORTS = ("newest", "oldest", "name", "name-desc", "rating")
+_LIBRARY_SORTS = ("newest", "oldest", "name", "rating")
+_MEDIA = ("all", "image", "video")
+_CONTROL = re.compile(r"[\x00-\x1f\x7f]")
+_SOURCE = re.compile(r"all|0|[1-9][0-9]{0,8}")
+FALLBACK_NAME = "Saved view"
+
+
+class SavedViewInvalid(ValueError):
+    """A request's page, name or state breaks the write rules (a 400)."""
+
+
+# (kind, limit-or-choices, default) per state key; "text" values are
+# length-capped. The defaults match the client's (saved-views.mjs FIELDS).
+_STATE_FIELDS = {
+    "images": {
+        "q": ("text", 500, ""),
+        "category": ("text", 100, ""),
+        "sort": ("choice", _IMAGE_SORTS, "newest"),
+        "source": ("source", None, "all"),
+        "folder": ("text", 1024, ""),
+        "media": ("choice", _MEDIA, "all"),
+        "tags": ("tags", None, ()),
+        # The Favorites page's filter (#249 D51): a view saved there reopens it.
+        "favorites": ("bool", None, False),
+    },
+    "library": {
+        "q": ("text", 500, ""),
+        "category": ("text", 100, ""),
+        "sort": ("choice", _LIBRARY_SORTS, "newest"),
+        "tags": ("tags", None, ()),
+    },
+}
+
+_DDL = (
+    f"CREATE TABLE {TABLE} (id INTEGER PRIMARY KEY AUTOINCREMENT, "
+    "page TEXT NOT NULL CHECK(page IN ('images','library')), "
+    "name TEXT NOT NULL, state TEXT NOT NULL, position INTEGER NOT NULL, "
+    "created_at TEXT NOT NULL, updated_at TEXT NOT NULL)"
+)
+_COLUMNS = (
+    ("id", "INTEGER", 0, 1),
+    ("page", "TEXT", 1, 0),
+    ("name", "TEXT", 1, 0),
+    ("state", "TEXT", 1, 0),
+    ("position", "INTEGER", 1, 0),
+    ("created_at", "TEXT", 1, 0),
+    ("updated_at", "TEXT", 1, 0),
+)
+_SELECT = "SELECT id, page, name, state FROM saved_views"
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql).lower()
+
+
+def validate_name(name: object) -> str:
+    """The trimmed name, 1 to 100 characters without control characters."""
+    if not isinstance(name, str):
+        raise SavedViewInvalid("Name must be a string")
+    trimmed = name.strip()
+    if not 1 <= len(trimmed) <= MAX_NAME_LENGTH:
+        raise SavedViewInvalid(f"Name must be 1 to {MAX_NAME_LENGTH} characters")
+    if _CONTROL.search(trimmed):
+        raise SavedViewInvalid("Name contains control characters")
+    return trimmed
+
+
+def _validate_value(key: str, kind: str, limit, value: object) -> None:
+    """Raise ``SavedViewInvalid`` when *value* breaks *key*'s write rule."""
+    if kind == "tags":
+        if not isinstance(value, list) or len(value) > MAX_TAGS:
+            raise SavedViewInvalid(
+                f"State '{key}' must be a list of at most {MAX_TAGS}"
+            )
+        for tag in value:
+            if (
+                not isinstance(tag, str)
+                or not tag.strip()
+                or len(tag) > MAX_TAG_LENGTH
+                or "," in tag
+                or _CONTROL.search(tag)
+            ):
+                raise SavedViewInvalid(f"State '{key}' holds an invalid tag")
+        return
+    if kind == "bool":
+        if not isinstance(value, bool):
+            raise SavedViewInvalid(f"State '{key}' must be true or false")
+        return
+    if not isinstance(value, str):
+        raise SavedViewInvalid(f"State '{key}' must be a string")
+    if kind == "choice" and value not in limit:
+        raise SavedViewInvalid(f"State '{key}' is not a supported value")
+    if kind == "source" and not _SOURCE.fullmatch(value):
+        raise SavedViewInvalid(f"State '{key}' must be 'all' or a source index")
+    if kind == "text" and (len(value) > limit or _CONTROL.search(value)):
+        raise SavedViewInvalid(f"State '{key}' is too long or has control characters")
+
+
+def encode_state(page: object, state: object) -> str:
+    """Validate *state* for *page* and return its compact JSON (at most 4 KB)."""
+    if page not in PAGES:
+        raise SavedViewInvalid("Page must be 'images' or 'library'")
+    fields = _STATE_FIELDS[page]
+    if not isinstance(state, dict):
+        raise SavedViewInvalid("State must be an object")
+    if set(state) != set(fields):
+        raise SavedViewInvalid(
+            f"State for {page} must contain exactly " + ", ".join(sorted(fields))
+        )
+    for key, (kind, limit, _default) in fields.items():
+        _validate_value(key, kind, limit, state[key])
+    encoded = json.dumps(
+        {key: state[key] for key in fields}, ensure_ascii=False, separators=(",", ":")
+    )
+    if len(encoded.encode("utf-8")) > MAX_STATE_BYTES:
+        raise SavedViewInvalid("State must encode to at most 4 KB")
+    return encoded
+
+
+def _decode_object(raw: object) -> Optional[dict]:
+    """The JSON object stored in *raw*, or None when it holds no object."""
+    if isinstance(raw, dict):
+        return raw
+    if not isinstance(raw, str):
+        return None
+    try:
+        decoded = json.loads(raw)
+    except (ValueError, RecursionError):
+        return None
+    return decoded if isinstance(decoded, dict) else None
+
+
+def normalize_state(page: str, raw: object) -> dict:
+    """The page's full state from a stored row, never raising.
+
+    Every key of the page's current whitelist is present: a stored value that
+    passes today's write rule is kept, anything else (missing, invalid, or a
+    row that is not a JSON object) takes the key's default, and keys the
+    whitelist no longer has are dropped. An unknown page has no keys.
+    """
+    fields = _STATE_FIELDS.get(page, {})
+    stored = _decode_object(raw) or {}
+    state = {}
+    for key, (kind, limit, default) in fields.items():
+        value = stored.get(key, default)
+        try:
+            _validate_value(key, kind, limit, value)
+        except SavedViewInvalid:
+            value = default
+        state[key] = list(value) if kind == "tags" else value
+    return state
+
+
+def normalize_name(name: object) -> str:
+    """A stored name made displayable: trimmed, capped, never empty."""
+    text = _CONTROL.sub("", name) if isinstance(name, str) else ""
+    trimmed = text.strip()[:MAX_NAME_LENGTH].strip()
+    return trimmed or FALLBACK_NAME
+
+
+def saved_views_schema_error(
+    conn: sqlite3.Connection,
+    *,
+    allow_missing: bool = True,
+    check_states: bool = True,
+) -> Optional[str]:
+    """Reject unsupported saved-views schemas and structurally invalid rows.
+
+    Reads only. Checks the table, its columns, each row's types and sizes and
+    that each state is a JSON object; never the value rules, which change
+    between versions and are normalized on read instead. A missing table is
+    acceptable for backups that predate it unless ``allow_missing`` is false.
+    ``check_states=False`` (live startup) skips the JSON-object check; such
+    rows are reported by ``unusable_view_ids`` and skipped on read.
+    """
+    objects = conn.execute(
+        "SELECT type,name,sql FROM sqlite_master WHERE name=? "
+        "OR (type IN ('index','trigger') AND tbl_name=?)",
+        (TABLE, TABLE),
+    ).fetchall()
+    if not objects:
+        return None if allow_missing else "Missing saved views schema"
+    tables = [sql for kind, _name, sql in objects if kind == "table"]
+    if len(tables) != 1 or len(objects) != 1:
+        return "Saved views name is reserved by an unsupported schema object"
+    if _compact(tables[0] or "") != _compact(_DDL):
+        return "Saved views has an invalid table definition"
+    columns = tuple(
+        (row[1], row[2].upper(), row[3], row[5])
+        for row in conn.execute(f"PRAGMA table_info({TABLE})")
+    )
+    if columns != _COLUMNS:
+        return "Saved views has invalid columns"
+    # Literal SQL: the table name and limits are constants of this module.
+    if conn.execute(
+        "SELECT 1 FROM saved_views WHERE typeof(page) != 'text' "
+        "OR page NOT IN ('images','library') "
+        "OR typeof(name) != 'text' OR typeof(state) != 'text' "
+        "OR typeof(position) != 'integer' OR typeof(created_at) != 'text' "
+        "OR typeof(updated_at) != 'text' "
+        "OR length(CAST(name AS BLOB)) > 400 "
+        "OR length(CAST(state AS BLOB)) > 4096 "
+        "OR length(CAST(created_at AS BLOB)) > 64 "
+        "OR length(CAST(updated_at AS BLOB)) > 64 LIMIT 1"
+    ).fetchone():
+        return "Saved views contains invalid values"
+    if check_states and unusable_view_ids(conn):
+        return "Saved views contains invalid values"
+    return None
+
+
+def unusable_view_ids(conn: sqlite3.Connection) -> list[int]:
+    """Ids of rows whose state is not a JSON object (never listed)."""
+    return [
+        view_id
+        for view_id, state in conn.execute("SELECT id, state FROM saved_views")
+        if _decode_object(state) is None
+    ]
+
+
+def migrate_saved_views(conn: sqlite3.Connection) -> None:
+    """Install the table atomically; validate before and after creating it."""
+    conn.execute("SAVEPOINT saved_views_schema")
+    try:
+        error = saved_views_schema_error(conn, check_states=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute(_DDL.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        error = saved_views_schema_error(conn, allow_missing=False, check_states=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        # A corrupt state never stops the live database from opening: the row
+        # stays (no data is destroyed) and reads skip it. Only its id is logged.
+        for view_id in unusable_view_ids(conn):
+            _LOGGER.warning(
+                "Saved view id %d has an unreadable state and is hidden", view_id
+            )
+        conn.execute("RELEASE saved_views_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO saved_views_schema")
+        conn.execute("RELEASE saved_views_schema")
+        raise
+
+
+def _view(row) -> dict:
+    view_id, page, name, state = row
+    return {
+        "id": view_id,
+        "page": page,
+        "name": normalize_name(name),
+        "state": normalize_state(page, state),
+    }
+
+
+class SavedViewOperations:
+    """PromptDatabase mixin: list, create, rename and delete saved views."""
+
+    def list_saved_views(self) -> list[dict]:
+        """Every view, in list order (position, then creation)."""
+        rows = self.model.get_connection().execute(_SELECT + " ORDER BY position, id")
+        # A row whose state is no JSON object (logged at startup) is never listed.
+        return [_view(row) for row in rows if _decode_object(row[3]) is not None]
+
+    def create_saved_view(self, page: object, name: object, state: object) -> dict:
+        """Validate and append one view at the end of the list."""
+        from .models import utc_now_iso  # models imports this module's schema
+
+        encoded = encode_state(page, state)
+        trimmed = validate_name(name)
+        now = utc_now_iso()
+        with self.model.get_connection() as conn:
+            cursor = conn.execute(
+                "INSERT INTO saved_views (page, name, state, position, "
+                "created_at, updated_at) SELECT ?, ?, ?, "
+                "COALESCE(MAX(position), -1) + 1, ?, ? FROM saved_views",
+                (page, trimmed, encoded, now, now),
+            )
+            row = conn.execute(
+                _SELECT + " WHERE id = ?", (cursor.lastrowid,)
+            ).fetchone()
+        return _view(row)
+
+    def rename_saved_view(self, view_id: int, name: object) -> Optional[dict]:
+        """The renamed view, or None when no view has *view_id*."""
+        from .models import utc_now_iso
+
+        trimmed = validate_name(name)
+        with self.model.get_connection() as conn:
+            conn.execute(
+                "UPDATE saved_views SET name = ?, updated_at = ? WHERE id = ?",
+                (trimmed, utc_now_iso(), view_id),
+            )
+            row = conn.execute(_SELECT + " WHERE id = ?", (view_id,)).fetchone()
+        # A hidden (unreadable) view is not found, as in the listing.
+        return _view(row) if row and _decode_object(row[3]) is not None else None
+
+    def delete_saved_view(self, view_id: int) -> bool:
+        """True when a view was removed."""
+        with self.model.get_connection() as conn:
+            cursor = conn.execute("DELETE FROM saved_views WHERE id = ?", (view_id,))
+        return cursor.rowcount > 0

database/vocabulary.py

@@ -0,0 +1,461 @@
+"""Bounded tag and category reads, and guarded tag commands.
+
+A count is the number of eligible Library parents that carry the value
+directly or through one active saved pair, which is exactly what
+``list_library(tags=[name])`` / ``list_library(category=value)`` return.
+A tag row's ``image_count`` is the number of files (private locators) with any
+byte record carrying the tag (#249), so a file counts once however many of its
+byte versions carry it, and equal bytes at two paths count twice.
+
+Tag commands run in one ``_pair_transaction``: the storage epoch is checked
+first, then the exact ``{id, name}`` identity, and the receipt is read on the
+same connection before commit. Owner rows and their timestamps never change.
+
+Category consolidation is previewed in one ``library_snapshot`` and applied in
+one ``_pair_transaction`` that recomputes the owner projection and requires the
+preview's ``impact_token``; only rows carrying the exact source change.
+"""
+
+import sqlite3
+
+from .library import _ELIGIBLE, _limit
+from .library_commands import _TOKEN, LibraryCommandError, _validate_epoch
+from .library_tokens import canonical_token
+from .models import utc_now_iso
+
+try:
+    from ..utils.validators import validate_category, validate_tags
+except ImportError:
+    from utils.validators import validate_category, validate_tags
+
+_CATEGORY_DOMAIN = "prompt-manager-vocabulary-category:v1\n"
+# Files carrying a tag through any of their byte records (#249 image tags).
+_IMAGE_FILES = (
+    "SELECT mit.tag_id,COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
+    "JOIN media_annotation_bytes b ON b.id=mit.byte_id"
+)
+
+
+def _tag_rows(page: str) -> str:
+    """Count rows for the registry rows selected by ``page`` (id, name)."""
+    # Eligible parent components carrying a tag in the page, plus the
+    # positive parent of every active pair carrying it (always eligible).
+    return (
+        "WITH page AS (" + page + "),"
+        "owners(tag_id,pid) AS ("
+        "SELECT pt.tag_id,p.id FROM prompts p "
+        "JOIN prompt_tags pt ON pt.prompt_id=p.id "
+        "WHERE pt.tag_id IN (SELECT id FROM page) AND " + _ELIGIBLE + " "
+        "UNION SELECT ppt.tag_id,pp.positive_prompt_id FROM prompt_pairs pp "
+        "JOIN prompt_pair_tags ppt ON ppt.pair_id=pp.id "
+        "WHERE pp.archived=0 AND ppt.tag_id IN (SELECT id FROM page)),"
+        "images(tag_id,n) AS (" + _IMAGE_FILES + " "
+        "WHERE mit.tag_id IN (SELECT id FROM page) GROUP BY mit.tag_id) "
+        "SELECT page.id,page.name,COUNT(DISTINCT owners.pid) AS prompt_count,"
+        "COALESCE(MAX(images.n),0) AS image_count "
+        "FROM page LEFT JOIN owners ON owners.tag_id=page.id "
+        "LEFT JOIN images ON images.tag_id=page.id "
+        "GROUP BY page.id ORDER BY page.id"
+    )
+
+
+_TAG_PAGE = _tag_rows("SELECT id,name FROM tags WHERE id>? ORDER BY id LIMIT ?")
+_TAG_ONE = _tag_rows("SELECT id,name FROM tags WHERE id=?")
+# Every owner kind a tag can sit on, for the delete confirmation.
+_TAG_IMPACT = (
+    "SELECT "
+    "(SELECT COUNT(*) FROM prompts p JOIN prompt_tags pt ON pt.prompt_id=p.id "
+    "WHERE pt.tag_id=:tag AND " + _ELIGIBLE + ") AS positive_prompts,"
+    "(SELECT COUNT(*) FROM prompts p JOIN prompt_tags pt ON pt.prompt_id=p.id "
+    "WHERE pt.tag_id=:tag AND NOT " + _ELIGIBLE + ") AS other_prompts,"
+    "(SELECT COUNT(*) FROM prompt_pairs pp JOIN prompt_pair_tags ppt "
+    "ON ppt.pair_id=pp.id WHERE ppt.tag_id=:tag AND pp.archived=0) "
+    "AS negative_variants,"
+    "(SELECT COUNT(*) FROM prompt_pairs pp JOIN prompt_pair_tags ppt "
+    "ON ppt.pair_id=pp.id WHERE ppt.tag_id=:tag AND pp.archived!=0) "
+    "AS archived_negative_variants,"
+    "(SELECT COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
+    "JOIN media_annotation_bytes b ON b.id=mit.byte_id WHERE mit.tag_id=:tag) "
+    "AS images"
+)
+_MAX_ID = 2**63 - 1
+
+
+def validate_vocabulary_tag_name(value: object) -> str:
+    """Return the trimmed name, applying tag rules plus the native comma ban."""
+    if not isinstance(value, str):
+        raise ValueError("Tag name must be a string")
+    validate_tags([value])
+    name = value.strip()
+    if "," in name:
+        raise ValueError("Tag names cannot contain commas")
+    return name
+
+
+def _validate_tag_id(tag_id: object) -> None:
+    if type(tag_id) is not int or not 1 <= tag_id <= _MAX_ID:
+        raise ValueError("Invalid tag ID")
+
+
+# The /library/options universe: nonblank values on eligible parents or on
+# active pairs, grouped exactly (whitespace and case variants stay distinct).
+_CATEGORY_GROUPS = (
+    "WITH owners(value,pid) AS ("
+    "SELECT p.category,p.id FROM prompts p WHERE p.category IS NOT NULL "
+    "AND TRIM(p.category)!='' AND " + _ELIGIBLE + " "
+    "UNION SELECT pp.category,pp.positive_prompt_id FROM prompt_pairs pp "
+    "WHERE pp.archived=0 AND pp.category IS NOT NULL AND TRIM(pp.category)!=''),"
+    "grouped AS (SELECT value,MIN(pid) AS first_pid,"
+    "COUNT(DISTINCT pid) AS prompt_count FROM owners GROUP BY value) "
+)
+
+
+# Every owner of one exact category value, in id order, with the flag the
+# impact counts split on.
+_CATEGORY_COMPONENTS = (
+    "SELECT p.id," + _ELIGIBLE + " AS eligible FROM prompts p "
+    "WHERE p.category=? ORDER BY p.id"
+)
+_CATEGORY_PAIRS = "SELECT id,archived FROM prompt_pairs WHERE category=? ORDER BY id"
+_CATEGORY_EXISTS = (
+    "SELECT EXISTS(SELECT 1 FROM prompts WHERE category=:v)"
+    " OR EXISTS(SELECT 1 FROM prompt_pairs WHERE category=:v)"
+)
+
+# Pairs after components, so a failing pair update rolls both back.
+_CATEGORY_UPDATES = (
+    "UPDATE prompts SET category=?,updated_at=? WHERE category=?",
+    "UPDATE prompt_pairs SET category=?,updated_at=? WHERE category=?",
+)
+
+
+def category_impact_token(
+    epoch: str,
+    source: str,
+    target: str,
+    target_exists: bool,
+    components: list,
+    pairs: list,
+) -> str:
+    """Hash a category projection like ``library_edit_token`` (no secret)."""
+    preimage = {
+        "epoch": epoch,
+        "source": source,
+        "target": target,
+        "target_exists": bool(target_exists),
+        "components": sorted([int(i), bool(flag)] for i, flag in components),
+        "pairs": sorted([int(i), bool(flag)] for i, flag in pairs),
+    }
+    return canonical_token(_CATEGORY_DOMAIN, preimage)
+
+
+def _validate_category_change(source: object, target: object) -> str:
+    """Check the exact source and return the trimmed, different target."""
+    if not isinstance(source, str) or not source.strip():
+        raise ValueError("source must be a nonblank category value")
+    if not isinstance(target, str):
+        raise ValueError("Category must be a string")
+    validate_category(target)
+    target = target.strip()
+    if not target:
+        raise ValueError("New category name cannot be blank")
+    if target == source:
+        raise ValueError("New category name must differ from the current one")
+    return target
+
+
+def _validate_impact_token(impact_token: object) -> None:
+    if not isinstance(impact_token, str) or not _TOKEN.fullmatch(impact_token):
+        raise ValueError("impact_token must be 64 lowercase hexadecimal characters")
+
+
+def _tag_after(after: object) -> int:
+    if after is None:
+        return 0
+    if type(after) is not int or after <= 0:
+        raise ValueError("Invalid vocabulary cursor")
+    return after
+
+
+def _category_after(after: object) -> tuple[int, str] | None:
+    if after is None:
+        return None
+    if (
+        not isinstance(after, list)
+        or len(after) != 2
+        or type(after[0]) is not int
+        or not isinstance(after[1], str)
+    ):
+        raise ValueError("Invalid vocabulary cursor")
+    return after[0], after[1]
+
+
+class VocabularyOperations:
+    """Vocabulary reads and tag commands.
+
+    ``_`` helpers take an open ``library_snapshot`` or ``_pair_transaction``
+    connection.
+    """
+
+    def _vocabulary_tag(self, conn: sqlite3.Connection, tag_id: int) -> dict | None:
+        row = conn.execute(_TAG_ONE, (tag_id,)).fetchone()
+        return dict(row) if row else None
+
+    def _vocabulary_identity(
+        self, conn: sqlite3.Connection, epoch: str, tag_id: int, name: str
+    ) -> dict:
+        """The stored tag, or vocabulary-conflict when it is gone or renamed."""
+        tag = self._vocabulary_tag(conn, tag_id)
+        if tag is None or tag["name"] != name:
+            raise LibraryCommandError("vocabulary-conflict", epoch)
+        return tag
+
+    @staticmethod
+    def _tag_owner_counts(conn: sqlite3.Connection, tag_id: int) -> dict:
+        components = conn.execute(
+            "SELECT COUNT(DISTINCT prompt_id) FROM prompt_tags WHERE tag_id=?",
+            (tag_id,),
+        ).fetchone()[0]
+        pairs = conn.execute(
+            "SELECT COUNT(DISTINCT pair_id) FROM prompt_pair_tags WHERE tag_id=?",
+            (tag_id,),
+        ).fetchone()[0]
+        images = conn.execute(
+            "SELECT COUNT(DISTINCT b.file_id) FROM media_image_tags mit "
+            "JOIN media_annotation_bytes b ON b.id=mit.byte_id WHERE mit.tag_id=?",
+            (tag_id,),
+        ).fetchone()[0]
+        return {
+            "components_changed": components,
+            "pairs_changed": pairs,
+            "images_changed": images,
+        }
+
+    def create_vocabulary_tag(self, name: str, *, expected_epoch: str) -> dict:
+        """Insert an unused registry row, or reuse the exact existing name."""
+        _validate_epoch(expected_epoch)
+        name = validate_vocabulary_tag_name(name)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            row = conn.execute("SELECT id FROM tags WHERE name=?", (name,)).fetchone()
+            created = row is None
+            if created:
+                tag_id = conn.execute(
+                    "INSERT INTO tags(name) VALUES(?)", (name,)
+                ).lastrowid
+            else:
+                tag_id = row[0]
+            return {
+                "library_epoch": epoch,
+                "created": created,
+                "tag": self._vocabulary_tag(conn, tag_id),
+            }
+
+    def rename_vocabulary_tag(
+        self, tag_id: int, *, expected_epoch: str, name: str, new_name: str
+    ) -> dict:
+        """Rename ``{tag_id, name}``, merging into an existing exact ``new_name``.
+
+        Links move across components and all pairs, archived included; an
+        owner already carrying the target keeps one link.
+        """
+        _validate_tag_id(tag_id)
+        _validate_epoch(expected_epoch)
+        if not isinstance(name, str):
+            raise ValueError("name must be the tag's current name")
+        new_name = validate_vocabulary_tag_name(new_name)
+        if new_name == name:
+            raise ValueError("New name must differ from the current name")
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            self._vocabulary_identity(conn, epoch, tag_id, name)
+            counts = self._tag_owner_counts(conn, tag_id)
+            target = conn.execute(
+                "SELECT id FROM tags WHERE name=?", (new_name,)
+            ).fetchone()
+            if target is None:
+                self._bump_image_tag_bytes(conn, tag_id)
+                conn.execute("UPDATE tags SET name=? WHERE id=?", (new_name, tag_id))
+                result_id = tag_id
+            else:
+                result_id = target[0]
+                self._move_tag_links(conn, tag_id, result_id)
+                conn.execute("DELETE FROM tags WHERE id=?", (tag_id,))
+            return {
+                "library_epoch": epoch,
+                "merged": target is not None,
+                "tag": self._vocabulary_tag(conn, result_id),
+                **counts,
+            }
+
+    def delete_vocabulary_tag(
+        self, tag_id: int, *, expected_epoch: str, name: str
+    ) -> dict:
+        """Remove the registry row ``{tag_id, name}``; its links cascade."""
+        _validate_tag_id(tag_id)
+        _validate_epoch(expected_epoch)
+        if not isinstance(name, str):
+            raise ValueError("name must be the tag's current name")
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            self._vocabulary_identity(conn, epoch, tag_id, name)
+            counts = self._tag_owner_counts(conn, tag_id)
+            self._bump_image_tag_bytes(conn, tag_id)
+            conn.execute("DELETE FROM tags WHERE id=?", (tag_id,))
+            return {"library_epoch": epoch, "deleted": True, **counts}
+
+    def vocabulary_tag_impact(self, tag_id: int) -> dict | None:
+        """Owner counts by kind for one tag, or None when it does not exist."""
+        _validate_tag_id(tag_id)
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            if not conn.execute("SELECT 1 FROM tags WHERE id=?", (tag_id,)).fetchone():
+                return None
+            row = conn.execute(_TAG_IMPACT, {"tag": tag_id}).fetchone()
+            return {"library_epoch": epoch, "impact": dict(row)}
+
+    @staticmethod
+    def _category_projection(
+        conn: sqlite3.Connection, epoch: str, source: str, target: str
+    ) -> dict:
+        """Owners of ``source``, impact counts and the token over them."""
+        components = [
+            [row[0], bool(row[1])]
+            for row in conn.execute(_CATEGORY_COMPONENTS, (source,))
+        ]
+        pairs = [
+            [row[0], bool(row[1])] for row in conn.execute(_CATEGORY_PAIRS, (source,))
+        ]
+        if not components and not pairs:
+            raise LibraryCommandError("vocabulary-conflict", epoch)
+        target_exists = bool(
+            conn.execute(_CATEGORY_EXISTS, {"v": target}).fetchone()[0]
+        )
+        eligible = sum(flag for _id, flag in components)
+        archived = sum(flag for _id, flag in pairs)
+        return {
+            "target_exists": target_exists,
+            "impact": {
+                "positive_prompts": eligible,
+                "other_prompts": len(components) - eligible,
+                "negative_variants": len(pairs) - archived,
+                "archived_negative_variants": archived,
+            },
+            "impact_token": category_impact_token(
+                epoch, source, target, target_exists, components, pairs
+            ),
+        }
+
+    def preview_vocabulary_category(
+        self, *, expected_epoch: str, source: str, target: str
+    ) -> dict:
+        """Read-only impact of renaming the exact ``source`` to ``target``."""
+        _validate_epoch(expected_epoch)
+        target = _validate_category_change(source, target)
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            if epoch != expected_epoch:
+                raise LibraryCommandError("storage-changed", epoch)
+            projection = self._category_projection(conn, epoch, source, target)
+            return {
+                "library_epoch": epoch,
+                "source": source,
+                "target": target,
+                **projection,
+            }
+
+    def apply_vocabulary_category(
+        self, *, expected_epoch: str, source: str, target: str, impact_token: str
+    ) -> dict:
+        """Set the exact ``source`` to ``target`` on every component and pair.
+
+        The recomputed projection must still match the preview's token; a
+        mismatch is ``vocabulary-conflict`` with nothing written.
+        """
+        _validate_epoch(expected_epoch)
+        target = _validate_category_change(source, target)
+        _validate_impact_token(impact_token)
+        with self._pair_transaction() as conn:
+            epoch = self._library_epoch(conn, expected_epoch)
+            projection = self._category_projection(conn, epoch, source, target)
+            if projection["impact_token"] != impact_token:
+                raise LibraryCommandError("vocabulary-conflict", epoch)
+            stamp = utc_now_iso()
+            for statement in _CATEGORY_UPDATES:
+                conn.execute(statement, (target, stamp, source))
+            return {
+                "library_epoch": epoch,
+                "source": source,
+                "target": target,
+                "consolidated": projection["target_exists"],
+                "impact": projection["impact"],
+            }
+
+    def _vocabulary_tags(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        *,
+        limit: int = 100,
+        after: int | None = None,
+    ) -> dict:
+        _limit(limit)
+        start = _tag_after(after)
+        total = conn.execute("SELECT COUNT(*) FROM tags").fetchone()[0]
+        rows = [dict(row) for row in conn.execute(_TAG_PAGE, (start, limit + 1))]
+        has_more = len(rows) > limit
+        rows = rows[:limit]
+        return {
+            "library_epoch": epoch,
+            "tags": rows,
+            "total": total,
+            "has_more": has_more,
+            "next_key": rows[-1]["id"] if has_more else None,
+        }
+
+    def _vocabulary_categories(
+        self,
+        conn: sqlite3.Connection,
+        epoch: str,
+        *,
+        limit: int = 100,
+        after: list | None = None,
+    ) -> dict:
+        _limit(limit)
+        key = _category_after(after)
+        total = conn.execute(
+            _CATEGORY_GROUPS + "SELECT COUNT(*) FROM grouped"
+        ).fetchone()[0]
+        where, params = "", []
+        if key is not None:
+            where = " WHERE first_pid>? OR (first_pid=? AND value>?)"
+            params = [key[0], key[0], key[1]]
+        rows = conn.execute(
+            _CATEGORY_GROUPS
+            + "SELECT value,first_pid,prompt_count FROM grouped"
+            + where
+            + " ORDER BY first_pid,value LIMIT ?",
+            [*params, limit + 1],
+        ).fetchall()
+        has_more = len(rows) > limit
+        rows = rows[:limit]
+        return {
+            "library_epoch": epoch,
+            "categories": [
+                {"value": row["value"], "prompt_count": row["prompt_count"]}
+                for row in rows
+            ],
+            "total": total,
+            "has_more": has_more,
+            "next_key": (
+                [rows[-1]["first_pid"], rows[-1]["value"]] if has_more else None
+            ),
+        }
+
+    def list_vocabulary_tags(self, *, limit: int = 100, after: int | None = None):
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            return self._vocabulary_tags(conn, epoch, limit=limit, after=after)
+
+    def list_vocabulary_categories(
+        self, *, limit: int = 100, after: list | None = None
+    ) -> dict:
+        with self.library_snapshot() as (conn, epoch, _ceiling):
+            return self._vocabulary_categories(conn, epoch, limit=limit, after=after)

database/workflow_documents.py

@@ -0,0 +1,652 @@
+"""Each workflow and prompt graph stored once, compressed (#249 S6 carry-over).
+
+``generated_images.workflow_data`` (the UI workflow) and ``prompt_metadata``
+(the API prompt graph) used to hold the same large JSON text again for every
+image of a run. Now each distinct text is one ``workflow_documents`` row:
+
+- ``sha256``: the SHA-256 of the text's exact UTF-8 bytes (unique key);
+- ``byte_length``: the length of those bytes;
+- ``data``: ``zlib.compress`` of those bytes.
+
+An image row references its texts through ``workflow_document_id`` and
+``prompt_document_id``. Two forms coexist until the migration has rewritten
+every row, and both stay readable forever:
+
+- old form: the text inline in ``workflow_data`` / ``prompt_metadata``, no id;
+- new form: the inline column NULL and the document id set.
+
+The text is stored and hashed exactly as written, never parsed and
+re-serialized, because a JSON round trip can change bytes (key order,
+spacing, number spelling, duplicate keys, ComfyUI's ``NaN`` tokens).
+
+``read_stored_text`` is the one decoder. Inline text, when present, wins (a
+row an older app version rewrote inline stays right); otherwise the document
+is read by its rowid, decompressed within its recorded length and checked
+against its length and hash. A document that fails a check reads as
+unreadable (None, logged), never as other text. ``resolve_image_row`` puts
+the texts back into a ``generated_images`` row and drops the id keys, so
+every payload keeps its shape.
+
+No trigger: every restore validator refuses triggers in a restored file. A
+document is removed in code: ``release_documents`` in the same transaction
+that drops a row's last reference, and ``sweep_documents`` (at startup, after
+the migration and before Reclaim space) for anything a path missed.
+"""
+
+import hashlib
+import logging
+import re
+import sqlite3
+import zlib
+from collections import OrderedDict
+from typing import Any, Dict, Iterable, Optional, Set
+
+logger = logging.getLogger("prompt_manager.database.workflow_documents")
+
+TABLE = "workflow_documents"
+# (inline text column, document id column) on generated_images.
+TEXT_COLUMNS = (
+    ("workflow_data", "workflow_document_id"),
+    ("prompt_metadata", "prompt_document_id"),
+)
+DOCUMENT_ID_COLUMNS = tuple(reference for _column, reference in TEXT_COLUMNS)
+INDEXES = {
+    "workflow_document_id": "idx_image_workflow_document",
+    "prompt_document_id": "idx_image_prompt_document",
+}
+COMPRESSION_LEVEL = 6
+# A text larger than this stays inline: a document is decompressed in one
+# piece, and its recorded length bounds that (a restored file cannot make a
+# small row expand without limit).
+MAX_DOCUMENT_BYTES = 256 * 2**20
+ID_BATCH = 400  # ids bound per statement (under SQLite's old 999 limit)
+
+_DDL = (
+    f"CREATE TABLE {TABLE} (id INTEGER PRIMARY KEY, sha256 BLOB NOT NULL UNIQUE, "
+    "byte_length INTEGER NOT NULL, data BLOB NOT NULL)"
+)
+_AUTOINDEX = f"sqlite_autoindex_{TABLE}_1"
+_REFERENCE_TYPE = f"INTEGER REFERENCES {TABLE}(id)"
+
+_UNREFERENCED = (
+    "NOT EXISTS (SELECT 1 FROM generated_images g "
+    "WHERE g.workflow_document_id=workflow_documents.id) "
+    "AND NOT EXISTS (SELECT 1 FROM generated_images g "
+    "WHERE g.prompt_document_id=workflow_documents.id)"
+)
+LOOKUP_SQL = (
+    "SELECT id, sha256, byte_length, data FROM workflow_documents WHERE sha256=?"
+)
+REPAIR_SQL = "UPDATE workflow_documents SET byte_length=?, data=? WHERE id=?"
+ALL_DOCUMENTS_SQL = "SELECT sha256, byte_length, data, id FROM workflow_documents"
+READ_SQL = "SELECT sha256, byte_length, data FROM workflow_documents WHERE id=?"
+INSERT_SQL = (
+    "INSERT INTO workflow_documents (sha256, byte_length, data) VALUES (?, ?, ?)"
+)
+# Built from constants only; the ids are bound as "?" marks.
+RELEASE_SQL = "DELETE FROM workflow_documents WHERE id IN ({marks}) AND "
+RELEASE_SQL += _UNREFERENCED
+# The sweep walks the ids through the small unique index, not the blobs.
+SWEEP_SQL = (
+    "DELETE FROM workflow_documents WHERE id IN (SELECT id FROM workflow_documents "
+    "INDEXED BY sqlite_autoindex_workflow_documents_1 WHERE "
+)
+SWEEP_SQL += _UNREFERENCED + ")"
+IMAGE_DOCUMENTS_SQL = (
+    "SELECT workflow_document_id, prompt_document_id FROM generated_images WHERE "
+)
+
+# Statements whose query plans tests/test_workflow_documents.py records.
+QUERY_PLAN_STATEMENTS = {
+    "lookup by hash": (LOOKUP_SQL, (b"\0" * 32,)),
+    "read by id": (READ_SQL, (1,)),
+    "release": (RELEASE_SQL.format(marks="?,?"), (1, 2)),
+    "sweep": (SWEEP_SQL, ()),
+    "documents of images": (IMAGE_DOCUMENTS_SQL + "id IN (?,?)", (1, 2)),
+    "documents of prompts": (IMAGE_DOCUMENTS_SQL + "prompt_id IN (?,?)", (1, 2)),
+}
+
+
+class DocumentMismatch(Exception):
+    """A document would not give back exactly the text it was made from."""
+
+
+def _compact(sql: str) -> str:
+    return re.sub(r"\s+", "", sql or "").lower()
+
+
+# ── schema ────────────────────────────────────────────────────────────
+
+
+def _reference_columns(conn: sqlite3.Connection) -> Dict[str, tuple]:
+    return {
+        row[1]: tuple(row)
+        for row in conn.execute("PRAGMA table_info(generated_images)")
+        if row[1] in DOCUMENT_ID_COLUMNS
+    }
+
+
+def _reference_error(conn: sqlite3.Connection, columns: Dict[str, tuple]):
+    for name, info in columns.items():
+        # (cid, name, type, notnull, default, pk)
+        if (str(info[2]).upper(), info[3], info[4], info[5]) != ("INTEGER", 0, None, 0):
+            return "Workflow document references have an invalid column"
+    keys = {
+        (row[3], row[2], row[4])
+        for row in conn.execute("PRAGMA foreign_key_list(generated_images)")
+        if row[3] in DOCUMENT_ID_COLUMNS
+    }
+    if keys != {(name, TABLE, "id") for name in DOCUMENT_ID_COLUMNS}:
+        return "Workflow document references have invalid foreign keys"
+    for column, index in INDEXES.items():
+        info = conn.execute(f'PRAGMA index_info("{index}")').fetchall()
+        owner = conn.execute(
+            "SELECT tbl_name FROM sqlite_master WHERE type='index' AND name=?",
+            (index,),
+        ).fetchone()
+        if [row[2] for row in info] != [column] or not owner:
+            return "Workflow document references are missing their index"
+        if owner[0] != "generated_images":
+            return "Workflow document references are missing their index"
+    return None
+
+
+def _value_error(conn: sqlite3.Connection):
+    if conn.execute(
+        "SELECT 1 FROM workflow_documents WHERE typeof(sha256)!='blob' "
+        "OR length(sha256)!=32 OR typeof(byte_length)!='integer' OR byte_length<0 "
+        "OR byte_length>? OR typeof(data)!='blob' LIMIT 1",
+        (MAX_DOCUMENT_BYTES,),
+    ).fetchone():
+        return "Workflow documents contain invalid values"
+    for column in DOCUMENT_ID_COLUMNS:
+        # typeof() reads no overflow page; a dangling id is a rowid probe.
+        # ``column`` is one of DOCUMENT_ID_COLUMNS, never caller data.
+        sql = (
+            f"SELECT 1 FROM generated_images g WHERE typeof(g.{column}) "  # nosec B608
+            f"NOT IN ('null','integer') OR (g.{column} IS NOT NULL AND NOT EXISTS "
+            f"(SELECT 1 FROM workflow_documents d WHERE d.id=g.{column})) LIMIT 1"
+        )
+        if conn.execute(sql).fetchone():
+            return "Workflow document references point at missing documents"
+    return None
+
+
+def damaged_message(document_id: int) -> str:
+    """The verify reason for a damaged document: which one, and what to do."""
+    return (
+        f"Workflow document {document_id} is damaged: its text no longer matches "
+        "its checksum, and the images that use it show no workflow. To fix it, "
+        "restore a backup from before the damage, or store a new image with the "
+        "same workflow (that repairs it)."
+    )
+
+
+def _damaged_document(conn: sqlite3.Connection) -> Optional[int]:
+    """The first document failing its length or hash check, or None."""
+    for row in conn.execute(ALL_DOCUMENTS_SQL):
+        if _decode(row[:3], row[3], quiet=True) is None:
+            return row[3]
+    return None
+
+
+def workflow_documents_schema_error(
+    conn: sqlite3.Connection, *, allow_missing: bool = True, documents: bool = False
+) -> Optional[str]:
+    """Reject an unsupported document schema or invalid stored values (reads only).
+
+    A database from before the dedupe has neither the table nor the id
+    columns; that is acceptable unless ``allow_missing`` is false. Half of
+    the schema never is.
+    """
+    objects = conn.execute(
+        "SELECT type,name,sql FROM sqlite_master WHERE name=? "
+        "OR (type IN ('index','trigger') AND tbl_name=?)",
+        (TABLE, TABLE),
+    ).fetchall()
+    has_images = conn.execute(
+        "SELECT 1 FROM sqlite_master WHERE type='table' AND name='generated_images'"
+    ).fetchone()
+    columns = _reference_columns(conn) if has_images else {}
+    if not objects and not columns:
+        if allow_missing or not has_images:
+            return None
+        return "Missing workflow document schema"
+    if not objects or set(columns) != set(DOCUMENT_ID_COLUMNS):
+        return "Incomplete workflow document schema"
+    tables = [sql for kind, _name, sql in objects if kind == "table"]
+    others = {(kind, name, sql) for kind, name, sql in objects if kind != "table"}
+    if len(tables) != 1 or others != {("index", _AUTOINDEX, None)}:
+        return "Workflow documents name is reserved by an unsupported schema object"
+    # The exact definition fixes the columns, their types and constraints.
+    if _compact(tables[0]) != _compact(_DDL):
+        return "Workflow documents have an invalid table definition"
+    error = _reference_error(conn, columns) or _value_error(conn)
+    if error is None and documents:
+        # Restore and backup verification only: every document is read.
+        damaged = _damaged_document(conn)
+        if damaged is not None:
+            return damaged_message(damaged)
+    return error
+
+
+def migrate_workflow_documents(conn: sqlite3.Connection) -> None:
+    """Install the table, the id columns and their indexes atomically.
+
+    Runs after every older migration (the table rebuilds only run on
+    databases without ``file_path``, before this). Validates before and
+    after, then sweeps documents no row references.
+    """
+    conn.execute("SAVEPOINT workflow_documents_schema")
+    try:
+        error = workflow_documents_schema_error(conn)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        conn.execute(_DDL.replace("CREATE TABLE", "CREATE TABLE IF NOT EXISTS", 1))
+        present = _reference_columns(conn)
+        for column in DOCUMENT_ID_COLUMNS:
+            if column not in present:
+                conn.execute(
+                    f"ALTER TABLE generated_images ADD COLUMN {column} "
+                    f"{_REFERENCE_TYPE}"
+                )
+        for column, index in INDEXES.items():
+            conn.execute(
+                f"CREATE INDEX IF NOT EXISTS {index} ON generated_images({column})"
+            )
+        error = workflow_documents_schema_error(conn, allow_missing=False)
+        if error:
+            raise sqlite3.DatabaseError(error)
+        sweep_documents(conn)
+        conn.execute("RELEASE workflow_documents_schema")
+    except BaseException:
+        conn.execute("ROLLBACK TO workflow_documents_schema")
+        conn.execute("RELEASE workflow_documents_schema")
+        raise
+
+
+# ── write ─────────────────────────────────────────────────────────────
+
+
+def storable(text: Any) -> bool:
+    """Whether ``text`` can be a document (text within MAX_DOCUMENT_BYTES)."""
+    if not isinstance(text, str):
+        return False
+    # UTF-8 takes at most 4 bytes a character: most texts need no encoding.
+    if len(text) * 4 <= MAX_DOCUMENT_BYTES:
+        return True
+    return len(text.encode("utf-8")) <= MAX_DOCUMENT_BYTES
+
+
+def store_text(
+    conn: sqlite3.Connection, text: Optional[str], verified: Optional[set] = None
+) -> Optional[int]:
+    """The id of the document holding exactly ``text``; None for None.
+
+    Reuses the document with the same hash, or inserts a new one whose
+    compressed bytes were checked to decompress to ``text``'s bytes. A reused
+    document is decompressed and checked first; a damaged one (a restored
+    file's corruption under a valid hash) is repaired from ``text``, whose
+    hash is its key, so every row that references it reads right again.
+    ``verified``, when given, holds the ids already checked (a migration
+    batch), which are then not read again. Opens a write transaction first
+    when none is open, so a concurrent ``release_documents`` cannot remove a
+    reused document before the caller references it. Raises TypeError for a
+    non-text value and ValueError for a text over MAX_DOCUMENT_BYTES.
+    """
+    if text is None:
+        return None
+    if not isinstance(text, str):
+        raise TypeError("Only text is stored as a workflow document")
+    raw = text.encode("utf-8")
+    if len(raw) > MAX_DOCUMENT_BYTES:
+        raise ValueError("Text is too large for a workflow document")
+    digest = hashlib.sha256(raw).digest()
+    if not conn.in_transaction:
+        conn.execute("BEGIN IMMEDIATE")
+    row = conn.execute(LOOKUP_SQL, (digest,)).fetchone()
+    if row is not None and verified is not None and row[0] in verified:
+        return row[0]
+    data = zlib.compress(raw, COMPRESSION_LEVEL)
+    if zlib.decompress(data) != raw:  # pragma: no cover - zlib is exact
+        raise DocumentMismatch("Compression did not round-trip")
+    if row is None:
+        document_id = conn.execute(INSERT_SQL, (digest, len(raw), data)).lastrowid
+    else:
+        document_id = row[0]
+        if _decode(row[1:], document_id, quiet=True) != text:
+            logger.warning(f"Workflow document {document_id} was damaged; repaired")
+            conn.execute(REPAIR_SQL, (len(raw), data, document_id))
+    if verified is not None:
+        verified.add(document_id)
+    return document_id
+
+
+def stored_values(
+    conn: sqlite3.Connection, texts: Dict[str, Optional[str]]
+) -> Dict[str, Any]:
+    """New-form column values for ``{text column: text}`` (writers).
+
+    Each text becomes ``{column: None, id column: document id}``; None
+    stays ``{column: None, id column: None}``; a value that cannot be a
+    document (not text, or too large) stays inline with no id.
+    """
+    references = dict(TEXT_COLUMNS)
+    values: Dict[str, Any] = {}
+    for column, text in texts.items():
+        reference = references[column]
+        if text is None or not storable(text):
+            values[column], values[reference] = text, None
+        else:
+            values[column], values[reference] = None, store_text(conn, text)
+    return values
+
+
+# ── read ──────────────────────────────────────────────────────────────
+
+
+def _decode(row, document_id, quiet=False) -> Optional[str]:
+    """``(sha256, byte_length, data)`` as checked text, or None (logged)."""
+    if row is None:
+        logger.warning(f"Workflow document {document_id} is missing")
+        return None
+    digest, length, data = bytes(row[0]), row[1], row[2]
+    try:
+        if not isinstance(length, int) or not 0 <= length <= MAX_DOCUMENT_BYTES:
+            raise ValueError("invalid length")
+        inflater = zlib.decompressobj()
+        raw = inflater.decompress(bytes(data), length + 1)
+        if len(raw) != length or inflater.unconsumed_tail or not inflater.eof:
+            raise ValueError("length differs")
+        if hashlib.sha256(raw).digest() != digest:
+            raise ValueError("hash differs")
+        return raw.decode("utf-8")
+    except (zlib.error, ValueError, TypeError) as error:
+        if not quiet:
+            logger.warning(f"Workflow document {document_id} is unreadable: {error}")
+        return None
+
+
+def read_stored_text(
+    conn: sqlite3.Connection, inline: Any, document_id: Optional[int]
+) -> Any:
+    """The stored text of one column: inline when present, else the document.
+
+    Returns None when there is neither, or when the document is missing or
+    fails its checks (logged). An inline value is returned unchanged.
+    """
+    if inline is not None:
+        return inline
+    if document_id is None:
+        return None
+    return _decode(conn.execute(READ_SQL, (document_id,)).fetchone(), document_id)
+
+
+class DocumentReader:
+    """``read_stored_text`` with a small memo for list reads on one connection."""
+
+    MEMO = 32
+
+    def __init__(self, conn: sqlite3.Connection):
+        self.conn = conn
+        self._memo: "OrderedDict[int, Optional[str]]" = OrderedDict()
+
+    def text(self, inline: Any, document_id: Optional[int]) -> Any:
+        if inline is not None or document_id is None:
+            return read_stored_text(self.conn, inline, document_id)
+        if document_id in self._memo:
+            self._memo.move_to_end(document_id)
+            return self._memo[document_id]
+        text = read_stored_text(self.conn, None, document_id)
+        self._memo[document_id] = text
+        if len(self._memo) > self.MEMO:
+            self._memo.popitem(last=False)
+        return text
+
+
+def resolve_image_row(reader: DocumentReader, row) -> Dict[str, Any]:
+    """A ``generated_images`` row as a dict with its texts read back.
+
+    The id columns are dropped, so a payload built from it has the shape it
+    had before the dedupe. Columns the row did not select stay absent.
+    """
+    data = dict(row)
+    for column, reference in TEXT_COLUMNS:
+        has_reference = reference in data
+        document_id = data.pop(reference, None)
+        if column in data or has_reference:
+            data[column] = reader.text(data.get(column), document_id)
+    return data
+
+
+# ── garbage ───────────────────────────────────────────────────────────
+
+
+def _chunks(values):
+    values = list(values)
+    for start in range(0, len(values), ID_BATCH):
+        yield values[start : start + ID_BATCH]
+
+
+def image_documents(conn: sqlite3.Connection, where: str, params) -> Set[int]:
+    """The document ids the image rows matching ``where`` reference.
+
+    ``where`` is a fixed condition written in this package (``id IN (...)``,
+    ``prompt_id IN (...)``), never caller data; values go in ``params``.
+    """
+    ids: Set[int] = set()
+    sql = IMAGE_DOCUMENTS_SQL + where
+    for row in conn.execute(sql, list(params)):  # nosec B608
+        ids.update(value for value in row if value is not None)
+    return ids
+
+
+def documents_of_rows(
+    conn: sqlite3.Connection, column: str, values: Iterable[Any]
+) -> Set[int]:
+    """Document ids of the image rows whose ``column`` is in ``values``.
+
+    ``column`` is ``id`` or ``prompt_id`` (the rows a prompt delete cascades to).
+    """
+    if column not in ("id", "prompt_id"):
+        raise ValueError(f"Unsupported column {column!r}")
+    ids: Set[int] = set()
+    for chunk in _chunks(values):
+        marks = ",".join("?" for _ in chunk)
+        ids |= image_documents(conn, f"{column} IN ({marks})", chunk)
+    return ids
+
+
+def release_documents(conn: sqlite3.Connection, ids: Iterable[Optional[int]]) -> int:
+    """Delete the documents in ``ids`` that no image row references now.
+
+    Runs in the caller's transaction, right after the delete or update that
+    dropped the references. Returns how many documents were removed.
+    """
+    removed = 0
+    for chunk in _chunks(sorted({i for i in ids if i is not None})):
+        marks = ",".join("?" for _ in chunk)
+        # Only "?" marks are interpolated.
+        sql = RELEASE_SQL.format(marks=marks)
+        removed += conn.execute(sql, chunk).rowcount  # nosec B608
+    return removed
+
+
+def sweep_documents(conn: sqlite3.Connection) -> int:
+    """Delete every document no image row references; returns the count."""
+    return conn.execute(SWEEP_SQL).rowcount
+
+
+# ── repair (a restore's stage) ────────────────────────────────────────
+
+INLINE_TEXTS_SQL = (
+    "SELECT workflow_data, prompt_metadata FROM generated_images "
+    "WHERE workflow_data IS NOT NULL OR prompt_metadata IS NOT NULL"
+)
+REKEY_SQL = "UPDATE workflow_documents SET sha256=?, byte_length=? WHERE id=?"
+
+
+def _rows(conn: sqlite3.Connection, sql: str, params=()):
+    """Each row of ``sql``, read one at a time.
+
+    Iterating a managed connection's cursor prefetches ``ITER_CHUNK`` (128)
+    rows, and a document or inline text can be megabytes.
+    """
+    cursor = conn.execute(sql, params)
+    while (row := cursor.fetchone()) is not None:
+        yield row
+
+
+def _inflated(data) -> Optional[bytes]:
+    """The complete text bytes ``data`` inflates to, or None.
+
+    Complete means the zlib stream ends (its Adler-32 checked) within
+    MAX_DOCUMENT_BYTES with nothing after it: the body is intact even when
+    the row's recorded hash or length is not.
+    """
+    try:
+        inflater = zlib.decompressobj()
+        raw = inflater.decompress(bytes(data), MAX_DOCUMENT_BYTES + 1)
+        if not inflater.eof or inflater.unconsumed_tail or inflater.unused_data:
+            return None
+        if len(raw) > MAX_DOCUMENT_BYTES:
+            return None
+        raw.decode("utf-8")
+        return raw
+    except (zlib.error, TypeError, UnicodeDecodeError):
+        return None
+
+
+def _one_column_off(row, raw: bytes) -> bool:
+    """Whether only the recorded hash or only the recorded length is wrong.
+
+    The two columns check the body independently. When the body inflates
+    completely and one of them still agrees with it, that column vouches for
+    the body and the other one is the damaged value. When both disagree, the
+    body itself is the suspect (another valid stream), so it is not trusted.
+    """
+    hash_ok = hashlib.sha256(raw).digest() == bytes(row[0])
+    length_ok = row[1] == len(raw)
+    return hash_ok != length_ok
+
+
+def _inline_copies(conn: sqlite3.Connection, wanted: Dict[bytes, int]):
+    """``{document id: text}`` for each wanted hash an inline row still holds."""
+    found: Dict[int, str] = {}
+    for row in _rows(conn, INLINE_TEXTS_SQL):
+        for text in row:
+            if not isinstance(text, str) or not storable(text):
+                continue
+            document_id = wanted.get(hashlib.sha256(text.encode("utf-8")).digest())
+            if document_id is not None and document_id not in found:
+                found[document_id] = text
+        if len(found) == len(wanted):
+            break
+    return found
+
+
+def _rekey(conn: sqlite3.Connection, document_id: int, raw: bytes) -> Optional[int]:
+    """Give an intact body its true key, or merge it into the document that has it.
+
+    Returns the holder's id when its own body was damaged and has just been
+    rewritten from ``raw`` (it is repaired, never to be cleared), else None.
+    """
+    digest = hashlib.sha256(raw).digest()
+    holder = conn.execute(LOOKUP_SQL, (digest,)).fetchone()
+    if holder is None or holder[0] == document_id:
+        conn.execute(REKEY_SQL, (digest, len(raw), document_id))
+        return None
+    for column in DOCUMENT_ID_COLUMNS:
+        # ``column`` is one of DOCUMENT_ID_COLUMNS, never caller data.
+        conn.execute(
+            f"UPDATE generated_images SET {column}=? WHERE {column}=?",  # nosec B608
+            (holder[0], document_id),
+        )
+    conn.execute("DELETE FROM workflow_documents WHERE id=?", (document_id,))
+    if _decode(holder[1:], holder[0], quiet=True) is None:
+        conn.execute(
+            REPAIR_SQL, (len(raw), zlib.compress(raw, COMPRESSION_LEVEL), holder[0])
+        )
+        return holder[0]
+    return None
+
+
+def repair_damaged_documents(conn: sqlite3.Connection) -> Dict[str, int]:
+    """Repair or clear every document failing its checks; commit (writes).
+
+    For a restore's stage only, never the live library. In order:
+
+    1. A document whose body inflates completely and whose recorded hash or
+       length (not both) disagrees is re-keyed from its text, or merged
+       into the document that already holds that text (``_one_column_off``).
+    2. A damaged body whose exact text an inline row still holds (the same
+       SHA-256, its key) is stored again from that text.
+    3. Any other damaged document is unreadable for good: its images'
+       references are cleared (they already showed no workflow) and the
+       document is removed, so the stage verifies.
+
+    Returns ``{repaired, cleared, images}``: documents repaired (1 and 2),
+    documents cleared and the distinct image rows whose reference was
+    cleared. Rows are read one at a time.
+    """
+    damaged: Dict[bytes, int] = {}
+    intact: Dict[int, bytes] = {}
+    for row in _rows(conn, ALL_DOCUMENTS_SQL):
+        if _decode(row[:3], row[3], quiet=True) is None:
+            raw = _inflated(row[2])
+            if raw is not None and _one_column_off(row, raw):
+                intact[row[3]] = raw
+            else:
+                damaged[bytes(row[0])] = row[3]
+    result = {"repaired": 0, "cleared": 0, "images": 0}
+    if not damaged and not intact:
+        return result
+    with conn:
+        healed = []
+        for document_id, raw in intact.items():
+            holder = _rekey(conn, document_id, raw)
+            if holder is not None and holder in damaged.values():
+                # Rewritten from the merged body: repaired, not to be cleared
+                # (it now also holds the merged images' references).
+                damaged = {k: v for k, v in damaged.items() if v != holder}
+                healed.append(holder)
+        copies = _inline_copies(conn, damaged) if damaged else {}
+        for document_id, text in copies.items():
+            raw = text.encode("utf-8")
+            data = zlib.compress(raw, COMPRESSION_LEVEL)
+            conn.execute(REPAIR_SQL, (len(raw), data, document_id))
+        lost = sorted(set(damaged.values()) - set(copies))
+        images: Set[int] = set()
+        for chunk in _chunks(lost):
+            marks = ",".join("?" * len(chunk))
+            # ``marks`` is only "?" placeholders; the ids are bound.
+            for (image_id,) in _rows(
+                conn,
+                "SELECT id FROM generated_images WHERE "  # nosec B608
+                f"workflow_document_id IN ({marks}) OR prompt_document_id IN ({marks})",
+                chunk + chunk,
+            ):
+                images.add(image_id)
+            for column in DOCUMENT_ID_COLUMNS:
+                conn.execute(
+                    f"UPDATE generated_images SET {column}=NULL "  # nosec B608
+                    f"WHERE {column} IN ({marks})",
+                    chunk,
+                )
+            conn.execute(
+                f"DELETE FROM workflow_documents WHERE id IN ({marks})",  # nosec B608
+                chunk,
+            )
+    for document_id in list(intact) + healed + list(copies):
+        logger.warning(f"Workflow document {document_id} was damaged; repaired")
+    if lost:
+        logger.warning(
+            f"{len(lost)} damaged workflow document(s) could not be repaired; "
+            f"{len(images)} image(s) show no workflow"
+        )
+    result.update(
+        repaired=len(intact) + len(healed) + len(copies),
+        cleared=len(lost),
+        images=len(images),
+    )
+    return result

database/workflow_storage.py

@@ -0,0 +1,455 @@
+"""The workflow storage migration and Reclaim space (#249 S6 carry-over).
+
+``migrate_workflow_storage`` rewrites old-form image rows (workflow and
+prompt graph text inline) to the new form (a ``workflow_documents`` id,
+``database/workflow_documents.py``) in batches:
+
+- each batch is one ``BEGIN IMMEDIATE`` transaction over at most
+  ``BATCH_ROWS`` rows or about ``BATCH_BYTES`` of text, found by a rowid
+  range search (``NEXT_IDS_SQL``) and read one row at a time, so memory is
+  bounded by one row plus the batch's distinct documents;
+- each text is stored (or its document reused) and read back through the
+  one accessor before its row changes; a text that would not come back
+  identical fails the batch (``DocumentMismatch``) and rolls it back whole;
+- a value that is not text (or is over ``MAX_DOCUMENT_BYTES``) stays inline
+  and is counted as protected;
+- each row runs in its own savepoint: one that cannot be rewritten (text
+  SQLite cannot decode, a text that would not come back) stays inline and
+  is listed, and the rest of the batch goes on;
+- a batch ends after ``BATCH_SECONDS``, and the write lock is then left
+  free for ``BATCH_PAUSE``, so other writers are never starved.
+
+A crash or a stop between batches leaves committed batches in the new form
+and the rest in the old form; both read the same, and a rerun continues
+from the first old-form row. Only database rows change: no image or video
+file is opened.
+
+``reclaim_space`` is the VACUUM afterwards: it sweeps unreferenced
+documents, rebuilds the file and checkpoints the WAL with TRUNCATE, so the
+freed pages go back to the disk.
+"""
+
+import logging
+import os
+import sqlite3
+import time
+from contextlib import closing
+
+from .workflow_documents import (
+    TEXT_COLUMNS,
+    DocumentMismatch,
+    read_stored_text,
+    release_documents,
+    storable,
+    store_text,
+    sweep_documents,
+)
+
+OPERATION = "workflow_storage"
+# The first apply's backup: ``<db>.pre-workflow-storage_<stamp>``, outside the
+# pruned ``<db>.backup_*`` pool. Only one is ever made; later applies take a
+# pooled safety backup.
+PRE_MIGRATION_LABEL = "pre-workflow-storage"
+BATCH_ROWS = 200
+BATCH_BYTES = 16 * 2**20  # characters of text rewritten per transaction
+# A batch holds the write lock: it ends after this long, and the loop then
+# leaves the lock free for BATCH_PAUSE, longer than SQLite's longest busy
+# poll (100 ms), so the image monitor and the run counting get in between.
+BATCH_SECONDS = 0.5
+BATCH_PAUSE = 0.15
+
+
+def _row_error(error) -> bool:
+    """Whether ``error`` is one row's own cause, which keeps only that row inline.
+
+    Text SQLite cannot decode for Python, and a text that would not come
+    back identical. Anything else (a full disk, an I/O error, a malformed
+    database, a bug) fails the whole batch. The decode error is matched by
+    its message: ``sqlite_errorcode`` needs Python 3.11.
+    """
+    if isinstance(error, (DocumentMismatch, UnicodeError)):
+        return True
+    return isinstance(error, sqlite3.OperationalError) and str(error).startswith(
+        "Could not decode"
+    )
+
+
+logger = logging.getLogger("prompt_manager.database.workflow_storage")
+_clock = time.monotonic
+
+# typeof() reads a column's type from the record header: no overflow page.
+NEXT_IDS_SQL = (
+    "SELECT id FROM generated_images WHERE id>? AND "
+    "(typeof(workflow_data)!='null' OR typeof(prompt_metadata)!='null') "
+    "ORDER BY id LIMIT ?"
+)
+ROW_SQL = (
+    "SELECT workflow_data, prompt_metadata, workflow_document_id, "
+    "prompt_document_id FROM generated_images WHERE id=?"
+)
+UPDATE_SQL = (
+    "UPDATE generated_images SET {assignments} WHERE id=?"  # columns are constants
+)
+# One pass over the record headers: rows with inline text, and inline
+# values that are not text (they stay inline).
+INLINE_COUNTS_SQL = (
+    "SELECT "
+    "SUM(typeof(workflow_data)='text' OR typeof(prompt_metadata)='text'), "
+    "SUM(typeof(workflow_data) NOT IN ('null','text')) "
+    "+ SUM(typeof(prompt_metadata) NOT IN ('null','text')) "
+    "FROM generated_images"
+)
+# octet_length() (SQLite 3.43+) reads no overflow page either.
+INLINE_BYTES_SQL = (
+    "SELECT SUM(octet_length(workflow_data)), SUM(octet_length(prompt_metadata)) "
+    "FROM generated_images"
+)
+
+QUERY_PLAN_STATEMENTS = {
+    "next old-form rows": (NEXT_IDS_SQL, (0, BATCH_ROWS)),
+    "one row": (ROW_SQL, (1,)),
+}
+
+
+def _rewrite_row(conn, image_id, row, verified, checked):
+    """Store the row's inline texts as documents; returns (changed, protected, size).
+
+    ``verified`` maps document ids already read back in this batch to their
+    text, and ``checked`` holds the reused ids ``store_text`` already
+    checked, so a document shared by many rows is decompressed once.
+    """
+    changes, replaced = {}, []
+    protected = size = 0
+    for index, (column, reference) in enumerate(TEXT_COLUMNS):
+        value, current = row[index], row[index + len(TEXT_COLUMNS)]
+        if value is None:
+            continue
+        if not storable(value):
+            protected += 1
+            continue
+        document_id = store_text(conn, value, checked)
+        if document_id not in verified:
+            verified[document_id] = read_stored_text(conn, None, document_id)
+        if verified[document_id] != value:
+            raise DocumentMismatch(f"Image {image_id}: {column} would change")
+        changes[column], changes[reference] = None, document_id
+        if current is not None and current != document_id:
+            replaced.append(current)
+        size += len(value)
+    if changes:
+        assignments = ",".join(f"{key}=?" for key in changes)
+        conn.execute(
+            UPDATE_SQL.format(assignments=assignments),
+            [*changes.values(), image_id],
+        )
+    release_documents(conn, replaced)
+    return bool(changes), protected, size
+
+
+def migrate_batch(
+    conn,
+    after_id,
+    max_rows=BATCH_ROWS,
+    max_bytes=BATCH_BYTES,
+    max_seconds=BATCH_SECONDS,
+):
+    """Rewrite the next batch after ``after_id`` in one transaction.
+
+    The batch ends after ``max_rows`` rows, about ``max_bytes`` of text or
+    ``max_seconds``, whichever comes first. Each row runs in its own
+    savepoint: a row that cannot be rewritten (``_row_error``) is rolled
+    back alone, stays inline and is listed in ``errors``; anything else rolls
+    the whole batch back and propagates. Returns ``{last_id, rows,
+    protected, errors, done}``: the last row looked at, rows rewritten,
+    inline values kept (not text), the ids kept by an error and whether
+    there was nothing left to look at.
+    """
+    if conn.in_transaction:
+        raise RuntimeError("The migration needs its own transaction")
+    started = _clock()
+    conn.execute("BEGIN IMMEDIATE")
+    try:
+        ids = [row[0] for row in conn.execute(NEXT_IDS_SQL, (after_id, max_rows))]
+        result = {
+            "last_id": after_id,
+            "rows": 0,
+            "protected": 0,
+            "errors": [],
+            "done": not ids,
+        }
+        verified, checked, size = {}, set(), 0
+        for image_id in ids:
+            conn.execute("SAVEPOINT workflow_row")
+            try:
+                row = conn.execute(ROW_SQL, (image_id,)).fetchone()
+                changed, protected, row_size = _rewrite_row(
+                    conn, image_id, row, verified, checked
+                )
+            except Exception as error:
+                if not _row_error(error):
+                    raise
+                conn.execute("ROLLBACK TO workflow_row")
+                conn.execute("RELEASE workflow_row")
+                verified.clear()  # the row's documents were rolled back too
+                checked.clear()
+                logger.warning(f"Image record {image_id} kept inline: {error}")
+                result["errors"].append(image_id)
+                changed, protected, row_size = False, 0, 0
+            else:
+                conn.execute("RELEASE workflow_row")
+            result["rows"] += int(changed)
+            result["protected"] += protected
+            result["last_id"] = image_id
+            size += row_size
+            if size >= max_bytes or _clock() - started >= max_seconds:
+                break
+        conn.commit()
+        return result
+    except BaseException:
+        conn.rollback()
+        raise
+
+
+def pre_migration_backups(db_path):
+    """The kept pre-migration backups beside ``db_path``, earliest first."""
+    db_path = os.path.abspath(db_path)
+    prefix = os.path.basename(db_path) + "." + PRE_MIGRATION_LABEL + "_"
+    folder = os.path.dirname(db_path)
+    try:
+        names = os.listdir(folder)
+    except OSError:
+        return []
+    return sorted(
+        name
+        for name in names
+        if name.startswith(prefix) and os.path.isfile(os.path.join(folder, name))
+    )
+
+
+def _file_bytes(db_path):
+    return sum(
+        os.path.getsize(path)
+        for path in (db_path, db_path + "-wal")
+        if os.path.exists(path)
+    )
+
+
+class WorkflowStorageOperations:
+    """PromptDatabase mixin: the workflow storage migration and Reclaim space."""
+
+    def migrate_workflow_storage(
+        self,
+        *,
+        on_batch=None,
+        should_stop=None,
+        max_rows=BATCH_ROWS,
+        max_bytes=BATCH_BYTES,
+    ) -> dict:
+        """Rewrite every old-form row, batch by batch (blocking).
+
+        ``on_batch(progress)`` runs after each committed batch with the
+        running totals ``{rows, protected, batches, last_id}``;
+        ``should_stop()`` is asked before each batch. Each committed batch is
+        followed by ``BATCH_PAUSE`` with the write lock free. Returns
+        ``{rows, protected, batches, stopped, errors, documents_removed}``
+        (``errors``: the ids of rows kept inline by an error). Unreferenced
+        documents are swept once every row was looked at.
+        """
+        totals = {"rows": 0, "protected": 0, "batches": 0, "last_id": 0}
+        errors = []
+        stopped = False
+        while True:
+            if should_stop is not None and should_stop():
+                stopped = True
+                break
+            batch = self.migrate_workflow_batch(totals["last_id"], max_rows, max_bytes)
+            if batch["done"]:
+                break
+            time.sleep(BATCH_PAUSE)  # leave the write lock to other writers
+            errors += batch["errors"]
+            totals["rows"] += batch["rows"]
+            totals["protected"] += batch["protected"]
+            totals["batches"] += 1
+            totals["last_id"] = batch["last_id"]
+            if on_batch is not None:
+                on_batch(dict(totals))
+        removed = 0 if stopped else self.sweep_workflow_documents()
+        return {
+            "rows": totals["rows"],
+            "protected": totals["protected"],
+            "batches": totals["batches"],
+            "stopped": stopped,
+            "errors": errors,
+            "documents_removed": removed,
+        }
+
+    def migrate_workflow_batch(
+        self, after_id, max_rows=BATCH_ROWS, max_bytes=BATCH_BYTES
+    ) -> dict:
+        """One batch on the calling thread's connection (``migrate_batch``)."""
+        return migrate_batch(self.model.get_connection(), after_id, max_rows, max_bytes)
+
+    def sweep_workflow_documents(self) -> int:
+        """Remove every document no image row references (one transaction)."""
+        conn = self.model.get_connection()
+        with conn:
+            return sweep_documents(conn)
+
+    def workflow_storage_status(self) -> dict:
+        """``workflow_storage_counts`` on one read-only snapshot (blocking)."""
+        from .maintenance import _read_snapshot
+
+        return _read_snapshot(self.model.db_path, self.workflow_storage_counts)
+
+    def workflow_storage_counts(self, conn) -> dict:
+        """``{rows, protected, documents, inline_bytes}`` on ``conn`` (reads only).
+
+        ``inline_bytes`` is None where SQLite has no ``octet_length``.
+        """
+        rows, protected = conn.execute(INLINE_COUNTS_SQL).fetchone()
+        documents = conn.execute("SELECT COUNT(*) FROM workflow_documents").fetchone()
+        inline_bytes = None
+        if sqlite3.sqlite_version_info >= (3, 43, 0):
+            row = conn.execute(INLINE_BYTES_SQL).fetchone()
+            inline_bytes = sum(value or 0 for value in row)
+        return {
+            "rows": rows or 0,
+            "protected": protected or 0,
+            "documents": documents[0],
+            "inline_bytes": inline_bytes,
+        }
+
+    def reclaim_space(self) -> dict:
+        """Sweep, VACUUM and truncate the WAL (blocking); file bytes before/after.
+
+        The caller holds the registry's ``exclusive`` slot. Raises whatever
+        stopped VACUUM; the database is then unchanged.
+        """
+        db_path = self.model.db_path
+        removed = self.sweep_workflow_documents()
+        before = _file_bytes(db_path)
+        self.model.vacuum_database()
+        with closing(sqlite3.connect(db_path)) as checkpoint:
+            busy = checkpoint.execute("PRAGMA wal_checkpoint(TRUNCATE)").fetchone()[0]
+        return {
+            "before_bytes": before,
+            "after_bytes": _file_bytes(db_path),
+            "documents_removed": removed,
+            "checkpoint_busy": bool(busy),
+        }
+
+    # ── preview ───────────────────────────────────────────────────────
+
+    def _plan_workflow_storage(self):
+        """The reviewed plan of the migration (one read-only snapshot)."""
+        from .maintenance import (
+            MaintenanceDraft,
+            _Fingerprint,
+            _megabytes,
+            _plural,
+            _read_snapshot,
+        )
+
+        def read(conn):
+            sizes = [
+                conn.execute(f"PRAGMA {name}").fetchone()[0]
+                for name in ("page_size", "page_count", "freelist_count")
+            ]
+            return sizes, self.workflow_storage_counts(conn)
+
+        (page_size, page_count, free_pages), counts = _read_snapshot(
+            self.model.db_path, read
+        )
+        size = page_size * page_count
+        details = [
+            {
+                "label": "Database file",
+                "detail": f"{_megabytes(size)}, of which "
+                f"{_megabytes(free_pages * page_size)} is free pages",
+            },
+            {
+                "label": "Image records",
+                "detail": f"{_plural(counts['rows'], 'record')} keep their workflow "
+                "or prompt graph inline; each distinct text is stored once, "
+                "compressed",
+            },
+        ]
+        if not counts["rows"]:
+            details[1]["detail"] = NOTHING_TO_DO
+        if counts["inline_bytes"] is not None:
+            details.append(
+                {
+                    "label": "Inline text",
+                    "detail": f"{_megabytes(counts['inline_bytes'])} now",
+                }
+            )
+        if counts["documents"]:
+            details.append(
+                {
+                    "label": "Already stored once",
+                    "detail": _plural(counts["documents"], "workflow document"),
+                }
+            )
+        name = os.path.basename(self.model.db_path)
+        kept = pre_migration_backups(self.model.db_path)
+        backup = (
+            POOLED_BACKUP_NOTE.format(
+                name=f"{name}.backup_<date and time>", kept=kept[0]
+            )
+            if kept
+            else BACKUP_NOTE.format(
+                name=f"{name}.{PRE_MIGRATION_LABEL}_<date and time>"
+            )
+        )
+        details += [
+            {"label": "Backup", "detail": backup},
+            {"label": "Reclaim space", "detail": RECLAIM_NOTE},
+            {"label": "Older versions", "detail": OLDER_VERSIONS_NOTE},
+        ]
+        protected = counts["protected"]
+        return MaintenanceDraft(
+            operation=OPERATION,
+            fingerprint=_Fingerprint(OPERATION).hexdigest(),
+            targets=[],
+            scope={
+                "count": counts["rows"],
+                "label": (
+                    _plural(counts["rows"], "image record") + " to rewrite"
+                    if counts["rows"]
+                    else "Nothing to rewrite"
+                ),
+            },
+            protected={
+                "count": protected,
+                "reasons": (
+                    [f"{_plural(protected, 'stored value')} not text; kept as is"]
+                    if protected
+                    else []
+                ),
+            },
+            media={"deletes_files": False, "label": MEDIA_NOTE},
+            details=details,
+        )
+
+
+MEDIA_NOTE = "No image or video file is opened or changed; only the database."
+RECLAIM_NOTE = (
+    "When every record is rewritten, Reclaim space runs alone to give the freed "
+    "space back to the disk (VACUUM). The database is locked while it runs."
+)
+BACKUP_NOTE = (
+    "Always taken first, whatever the safety backup setting: {name} beside "
+    "the database, verified, and kept until you remove it (it is never pruned)."
+)
+POOLED_BACKUP_NOTE = (
+    "A safety backup is taken first, whatever the safety backup setting: {name} "
+    "among the pooled backups; the pre-migration backup {kept} is already kept."
+)
+NOTHING_TO_DO = "Every image record already stores its workflow once."
+OLDER_VERSIONS_NOTE = (
+    "This version stores the workflows of images it records in the new form, "
+    "whether or not this job runs, and this job rewrites the rest. An older "
+    "Prompt Manager version shows those images without their workflow until you "
+    "update again. The way back to an older version is the earliest "
+    "pre-workflow-storage backup."
+)

docs/application-redesign.md

@@ -0,0 +1,85 @@
+# Application Redesign Development Plan
+
+The maintainer approved the [interactive concept](concepts/workspace/README.md)
+on 2026-10-17. It is a design reference with simulated operations, not production
+code. This plan proposes implementation order and acceptance gates.
+
+## Delivery status
+
+- Foundation merged in [#198](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/198).
+- [#200](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/200) delivers native Images browsing, physical-file details, infinite scroll, Grid/List/Masonry, Viewer.js and video playback. The workspace opens on Images; original application entry points remain available.
+- Shared role-aware sending and the physical-file prompt inspector merged in [#203](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/203) and [#206](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/206).
+- The native Prompt Library merged in [#210](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/210) for [#208](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/208): one positive parent, saved negative variants, full selected text and a shared filtered image gallery. It does not invent chronological runs from counters or scan dates.
+- [Durable history storage](prompt-history-storage.md) records job attempts and outcomes from [#212](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/212), frozen prompt contexts from [#213](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/213), and observed run-owned output locators from [#214](https://github.com/ComfyAssets/ComfyUI_PromptManager/pull/214). Those locators do not verify file bytes or establish gallery-image associations. That association contract and the chronological execution timeline remain pending under [#209](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/209).
+- [#215](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/215) brings the shell, Images and natural-height Prompt Library presentation into the approved geometry and themes while retaining their existing operations. [#216](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/216) adds the shared current-file inspector with exact generation facts, scoped provenance, raw/download evidence and loaded-collection viewer controls, including a slideshow that pauses at native video. This delivery does not establish whole-application parity, change the default entry point or constitute a release. Recorded chronological run/context browsing continues under [#209](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/209); image tagging/review, System consolidation and remaining organization capabilities continue under #196.
+- Local Inspect (#249, PRs L1–L2) replaces the `#/inspect` handoff with the concept's native "Inspect a file" page: one local image or video previews inline and opens the shared viewer and full inspector on a record read entirely in the browser, with nothing uploaded, imported or stored. The legacy metadata page (`web/metadata.html`, served at `/prompt_manager/gallery.html`) stays reachable from Overview; its capabilities map to the shared inspector, except its one-click COPY ALL summary, whose ruled replacement ("Copy settings summary") is pending. Recorded deviations D60–D66 are in [TESTING.md](../TESTING.md). Local roles equal to the server's resolver follow in PR L3.
+
+## Scope and compatibility
+
+Rebuild the standalone application around one shared shell. Preserve existing
+node schemas, saved workflows, backend capabilities, pinned Viewer.js and the
+ComfyAssets toolkit. Keep existing application entry points usable during the
+transition. Do not change the running maintainer instance during automated tests.
+
+Use the concept's feature inventory as a parity checklist. Each row needs an
+implementation issue, observable acceptance criteria and automated test coverage.
+Record any intentional behavior change explicitly; no silent feature removal.
+
+## Proposed delivery sequence
+
+Use a tracking issue and small issue-linked branches and PRs. Dependencies may
+require splitting these milestones further.
+
+| Milestone | Deliverable and acceptance focus |
+| --- | --- |
+| 1. Foundation | Production shell, routing, shared theme and disposable page-test harness. Verify startup, navigation, dark/light changes, cleanup and existing entry points. Establish discovered-test and coverage baselines before changes. |
+| 2. Image workspace | Real API-backed filters, Grid/List/Masonry, readable thumbnails, selection and shared Viewer.js inspector. Continuous infinite scroll only, with accessible Load more fallback, stable ordering, retry and restored browsing context. Bound retained DOM and test large datasets. |
+| 3. Prompt library and history | Real positive/negative associations, selected full negative, run history and shared image filtering. Design any missing execution-history persistence and migration before wiring the timeline; never invent historical runs from incomplete metadata. |
+| 4. Send to ComfyUI | Shared paper-plane actions wherever prompt text appears; role-specific and pair sends, destination preview, acknowledgement and failure recovery. Cover explicit empty versus unknown negatives and connected inputs. Never queue or replace a workflow. |
+| 5. Editing and review | Prompt editing, vocabulary management, per-image manual/auto-tag approval and Needs review. Preserve cancellation semantics and distinguish tag approval from prompt-association decisions. |
+| 6. System and maintenance | Settings, sources/scans, integrations, jobs, diagnostics, backup/restore and maintenance impact previews. Cover existing capabilities before retiring their old controls. |
+| 7. Additional organization | Favorites independent of ratings, collections, saved views, Activity, scheduled backups and portable import previews. Specify persistence, validation and recovery separately for each capability. |
+| 8. Cutover | Complete the parity checklist, migration and upgrade validation, accessibility/performance checks, documentation and targeted maintainer live testing. Switch the default only after these pass; preserve compatible URLs. |
+
+## TDD and test layers
+
+For each behavior, first write and run a meaningful failing test. Implement the
+smallest working change, verify it passes, then refactor with the suite green.
+Include failing-before/passing-after evidence in the PR.
+
+- **Unit:** filtering and selection state, cursor handling, prompt roles,
+  formatting, validation, navigation and lifecycle cleanup.
+- **Integration:** real HTTP handlers and disposable SQLite/filesystem fixtures;
+  persistence, migration from supported old schemas, idempotent scans, restore,
+  rollback and reference-safe deletion. Mock inference and external services.
+- **Smoke:** shipped production pages start and navigate with real fixture APIs;
+  test populated, empty and error states, assets and unexpected console failures.
+- **Browser workflows:** production HTML/JS across Chromium, Firefox, WebKit and
+  touch; editing/reload, infinite append, viewer cleanup, tagging approval,
+  keyboard access, delayed responses, cancellation and retries.
+- **Host integration:** pinned disposable CPU ComfyUI, Classic and Nodes 2.0,
+  switching both ways and workflow reload for host-facing changes. Exercise
+  independent and combined pack installations for shared bridge/toolkit changes.
+
+Retain existing test discovery and coverage gates. Carry forward database
+injection and open-handle checks into new fixtures. Add new frontend modules to
+coverage measurement and its baseline; do not lower floors or replace existing
+regression coverage with mockup-only tests.
+
+## GitHub and release gates
+
+Extend existing required workflows: the Python 3.10–3.13 matrix on Linux,
+Windows and macOS; 85% Python coverage; formatting; frontend units and per-file
+coverage floors; vendor checks; browser workflows; real ComfyUI compatibility.
+Publish traces, screenshots, logs and coverage artifacts on failures. See
+[TESTING.md](../TESTING.md) for commands and fixture isolation requirements.
+
+Every PR must map acceptance criteria to tests, report relevant checks and include
+screenshots for visual changes. All required checks must pass before merging.
+Use maintainer live review at meaningful visual and host-integration milestones;
+capture discovered regressions in automated tests. Notify the maintainer when
+Python changes require a ComfyUI restart.
+
+Keep version/publication changes separate from intermediate UI work: the existing
+registry workflow publishes on main/master changes to `pyproject.toml` or manual
+dispatch. Do not trigger a release merely to preview a redesign milestone.

docs/concepts/workspace/README.md

@@ -0,0 +1,138 @@
+# Prompt Manager application redesign concept
+
+Status: visual concept approved by the maintainer, 2026-10-17. Production
+implementation is progressing in tested slices; see the [development plan](../../application-redesign.md).
+The production workspace now has native Images browsing; this concept remains the
+approved target for the complete application.
+
+## Review the mockup
+
+Run `python tools/preview-prompt-timeline.py --bind 0.0.0.0 --port 8767` and
+open `/workspace/`. The original approved timeline remains available at `/`.
+The server exposes an explicit allowlist of concept files and pinned assets.
+All examples are synthetic; there are no application API calls, database access,
+real model downloads, or ComfyUI handoffs. Local file inspection previews an
+image in the browser without uploading it. Exports contain sample data only.
+
+The sidebar opens Images, Prompt library, Tags & categories, Inspect a file,
+Activity, System and Design review. System includes Overview, Appearance,
+Sources & scans, Integrations, Auto-tagging, Maintenance, Backup & export, and
+Diagnostics & logs. The in-app Design review contains the feature map below.
+
+## Proposed direction
+
+One application shell replaces separate Admin, Gallery and Metadata shells.
+Use the approved shared ComfyAssets theme: ComfyUI dark/light, blue primary
+actions, green positive roles, red negative roles and errors. Preserve existing
+capabilities and services while consolidating controls and clarifying their scope.
+
+Browsing is image-led with readable thumbnails and one metadata inspector.
+Prompt library groups by positive; negative variants and runs share its gallery.
+The full selected negative appears directly beneath the positive in history.
+Favorites express preference independently of the existing quality rating.
+Collections organize projects without duplicating recipe text; saved views keep
+search/filter context. New capabilities are proposals, not implemented features.
+
+Maintenance moves out of the browsing toolbar. Checks, impact preview, backups,
+and apply steps have separate responsibilities. Distinguish prompt text reuse
+from image-file duplicates, and metadata enrichment from thumbnail generation.
+No repeated global thumbnail-generation prompt on opening the gallery.
+
+## Send to ComfyUI
+
+Use the same original paper-plane outline, blue accent and accessible role label
+on every positive/negative prompt block, including image details and the viewer.
+A pair action sends both. Preview destination and exact fields before applying;
+positive-only and negative-only actions leave the other role untouched. An
+explicitly empty negative means clear the target field; unknown metadata must
+never be interpreted as empty. Sending text neither queues execution nor
+replaces a workflow.
+
+Production needs a shared host bridge with versioned payloads, origin validation,
+destination discovery, capability checks, disconnected-host handling, linked-input
+checks and acknowledgements. Preview actual target values and report success only
+after acknowledgement. Default to asking for a destination; never guess between
+multiple nodes or positive and negative encoder inputs. Verify Classic and Nodes
+2.0 with both packs installed independently and together.
+
+## Existing feature inventory
+
+| Current feature | Proposed home / behavior | Disposition |
+| --- | --- | --- |
+| Admin prompt cards / per-prompt gallery | Prompt library + one prompt history workspace | Retained, rebuilt |
+| Standalone image gallery / Admin image modal | One Images workspace + shared Viewer.js viewer | Consolidated |
+| Video playback, autoplay / mute / loop, viewer slideshow and transforms | Shared media viewer; playback controls in Appearance; same image navigation | Retained |
+| Metadata page / viewer sidebar / full-text popups | One reusable metadata inspector; inline complete prompt text | Consolidated |
+| Prompt add, edit, rating, notes, copy, delete | Recipe editor, shared details, contextual bulk actions; reference-safe archive | Retained, clarified |
+| Categories, tag rename / merge, autocomplete | Tags & categories; same vocabulary in every editor | Retained |
+| Grid/list, thumbnail size, columns, pages/infinite scroll | Infinite scroll only; one Appearance configuration and contextual layout / density overrides | Simplified |
+| Scan, rescan, monitoring directories, worker threads | Sources & scans; explicit metadata enrichment vs thumbnail work | Consolidated |
+| Ollama, HuggingFace, WD14 variants, thresholds, review | One Auto-tagging configuration and review flow | Retained |
+| Thumbnail quality, size limit, metadata cache, paths | Appearance + cache controls in Maintenance | Retained |
+| Prompt dedup, hash checks, consistency, unused prompts | Maintenance plan with dependency checks and impact preview | Retained, safer flow |
+| Image dedup, orphan image links, browser cache | Separate reviewed operations in Maintenance | Retained |
+| Database backup / restore, JSON export | Backup & export; recovery and portable recipes clearly distinguished | Retained |
+| Diagnostics, image-link checks, logs, rotation, log levels | System → Diagnostics & logs | Retained |
+| LoRA detection/import, trigger injection, CivitAI key | System → Integrations; positive-only triggers | Retained |
+| Statistics / counts | System overview + usage breakdown; context counts in the workspace | Retained |
+| Send prompts to ComfyUI from every prompt surface | Role-aware paper-plane action + destination / overwrite preview | New proposal |
+| Reusable negatives / approved branch timeline | Prompt history with selected negative directly beneath positive | Approved concept |
+| Favorites, saved searches, project collections | Sidebar views; collections do not duplicate recipe text | New proposal |
+| Ambiguous metadata review, source provenance, run comparisons | Needs review queue; recorded vs inferred evidence; pair/run selection | New / expanded |
+| Unified jobs, cancellation, resumable progress visibility | Activity center; cancellation distinguished from hiding progress | New proposal |
+| Scheduled backups, pre-change safety snapshots | Explicit recovery policy after storage configuration | New proposal |
+
+## Implementation boundaries and test requirements
+
+Keep Viewer.js 1.15.2 and the shared toolkit as reproducible pinned artifacts;
+this mock uses their actual checked-in files without modifying them. Preserve
+legacy routes during the transition, current node layouts/schemas, existing
+controllers where reusable, and all existing test/coverage gates.
+
+The timeline needs persisted execution records to represent real runs. Saved
+views, collections, metadata review resolution, activity history, scheduled
+backups, portable JSON import, and the ComfyUI bridge need separate implementation
+and persistence work. Preserve source evidence and distinguish recorded from
+inferred metadata. Existing schema migration is automatic; old image enrichment
+remains an optional explicit scan. The one-time upgrade notice belongs in the
+new shell, with Scan and Later choices.
+
+Implement in reviewed slices: shared application shell and inspector; reusable
+prompt blocks/send bridge; image workspace; grouped library/history; settings and
+maintenance; optional organization and workflow improvements. Write regressions
+first for production behavior. Retain Python, frontend unit, Chromium/Firefox/
+WebKit/touch and managed Classic/Nodes 2.0 tests. Add behavior tests for shared
+filters, navigation state, viewer cleanup, interrupted jobs, dependency-safe
+maintenance, backup/restore, paired metadata and correct send destinations.
+
+Screenshots in `screenshots/` record the reviewed starting point. The approved
+standalone timeline in `../prompt-variant-timeline.html` remains unchanged.
+
+## Review polish
+
+History actions use a shared branch icon. Images can switch among Grid, List and
+Masonry; Masonry preserves each image's proportions and keeps DOM order while
+packing cards with a responsive layout. It does not require experimental native
+masonry support. Resize observers are disconnected when leaving the view.
+
+Every image card includes a tagging action beside Info, including cards in Needs
+review. Choose WD14 SwinV2, WD14 ViT, Ollama or HuggingFace; adjust thresholds or
+instructions, generate sample suggestions, edit tags manually, and approve only
+the desired suggestions. Changes are staged until approval, apply to that image
+alone and do not resolve uncertain prompt associations. Changing models discards
+previous suggestions. Cancel preserves existing tags. All inference is simulated.
+
+## Continuous browsing decision
+
+The redesign uses infinite scroll only. Remove page numbers, previous/next page
+buttons, paged/infinite mode selection and results-per-page controls. Keep request
+batching internal. The mock appends 12 images at a time in Grid, List and Masonry;
+it preserves existing card nodes, focus and selection while appending. It also
+retains the loaded range when switching layouts or revisiting a search. A
+keyboard-accessible Load more action provides an alternate trigger for the same
+continuous stream, not pagination. Exhausted and empty results are explicit.
+
+Production will need bounded rendering, stable cursors, scroll restoration and
+loading/error/retry behavior for large collections. Existing backend pagination
+can continue providing internal request batches; no production tests or routes
+are removed as part of this prototype.

docs/concepts/workspace/index.html

@@ -0,0 +1,105 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width,initial-scale=1" />
+    <meta name="referrer" content="no-referrer" />
+    <title>Images · Prompt Manager concept</title>
+    <link rel="stylesheet" href="/toolkit/controls.css" />
+    <link rel="stylesheet" href="/vendor/viewer.css" />
+    <link rel="stylesheet" href="/workspace/style.css" />
+    <script src="/vendor/viewer.js" defer></script>
+    <script src="/workspace/app.js" type="module"></script>
+  </head>
+  <body class="ca-tools">
+    <a class="skip" href="#main">Skip to workspace</a>
+    <div class="app-shell">
+      <aside class="sidebar">
+        <a class="brand" href="#images"
+          ><span class="brand-mark">P<span>±</span></span
+          ><span>Prompt Manager<small>COMFYASSETS</small></span></a
+        >
+        <div class="sidebar-label">WORKSPACE</div>
+        <nav id="primary-nav" aria-label="Workspace"></nav>
+        <div class="sidebar-label">
+          SAVED VIEWS <span class="new-label">PROPOSED</span>
+        </div>
+        <nav id="saved-nav" aria-label="Saved views"></nav>
+        <div class="sidebar-label">
+          COLLECTIONS <span class="new-label">PROPOSED</span>
+        </div>
+        <nav id="collections-nav" aria-label="Collections"></nav>
+        <div class="sidebar-bottom">
+          <nav id="system-nav" aria-label="Tools and system"></nav>
+          <div class="connection">
+            <i></i>
+            <div>
+              ComfyUI workspace<small
+                >Demo connection · Classic / Nodes 2.0</small
+              >
+            </div>
+          </div>
+        </div>
+      </aside>
+      <div class="workspace">
+        <header class="topbar">
+          <div class="breadcrumbs">
+            ComfyAssets <span>/</span> <span id="breadcrumb">Images</span>
+          </div>
+          <div class="top-actions">
+            <span class="concept-badge">Interactive concept</span
+            ><button id="activity-open" type="button">
+              Activity <span class="count" id="activity-count">2</span></button
+            ><label class="theme-label" for="theme"
+              >Theme
+              <select id="theme" aria-label="Theme">
+                <option value="dark">Dark</option>
+                <option value="light">Light</option>
+                <option value="system">System</option>
+              </select></label
+            ><a class="canvas-link" href="#send"
+              >ComfyUI <span aria-hidden="true">↗</span></a
+            >
+          </div>
+        </header>
+        <main id="main" tabindex="-1"></main>
+        <footer class="app-footer">
+          <span
+            >Design prototype · sample images and data · actions stay in this
+            tab</span
+          ><button type="button" id="review-link">What changes & why</button>
+        </footer>
+      </div>
+    </div>
+    <div id="toast" class="toast" role="status" aria-atomic="true"></div>
+    <dialog id="modal" aria-labelledby="modal-title">
+      <div class="dialog-header">
+        <h2 id="modal-title"></h2>
+        <button type="button" id="modal-close" aria-label="Close dialog">
+          ×
+        </button>
+      </div>
+      <div id="modal-body"></div>
+    </dialog>
+    <dialog id="viewer-dialog" aria-labelledby="viewer-title">
+      <div class="dialog-header">
+        <div>
+          <h2 id="viewer-title">Image details</h2>
+          <p class="muted small">
+            One viewer · image, prompts, generation data, workflow
+          </p>
+        </div>
+        <button type="button" id="viewer-close" aria-label="Close image viewer">
+          ×
+        </button>
+      </div>
+      <div class="viewer-layout">
+        <div class="viewer-left">
+          <div id="viewer-stage"></div>
+          <div class="viewer-toolbar" id="viewer-toolbar"></div>
+        </div>
+        <aside id="viewer-details" aria-label="Image metadata"></aside>
+      </div>
+    </dialog>
+  </body>
+</html>

docs/prompt-history-storage.md

@@ -0,0 +1,84 @@
+# Durable Prompt History
+
+Status: the durable job ledger, immutable prompt contexts and run-owned output
+references are implemented for
+[#209](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/209). Gallery
+image associations and the chronological timeline remain pending; #209 stays open.
+
+## Recorded jobs
+
+New jobs observed at the ComfyUI queue's dequeue boundary receive an immutable
+run identity and a host-session/queue-item identity. The public ComfyUI job UUID
+is a correlation field: submitting it twice records two attempts. Rejected graphs
+and jobs removed from the queue before dequeue create no execution rows. Cached
+jobs are recorded even when no PromptManager node runs, and headless jobs do not
+require a browser or WebSocket client.
+
+The additive `execution_runs` table stores dequeue time, observed callback time,
+optional executor event times and an explicit outcome: `observed`, `succeeded`,
+`failed`, `interrupted` or `unknown`. Matching stored terminal events determine
+outcomes. Missing or conflicting terminal evidence remains unknown; a process
+crash can leave an observed row without an invented completion time. Bounded
+safe diagnostics retain node ID/type and exception type, never graph text,
+arbitrary extra data, credentials, exception messages, traceback, inputs or
+arbitrary output payloads. A bounded background writer keeps storage failures from
+changing host execution and reports capture gaps rather than inventing rows.
+
+Existing databases begin with zero execution rows. Migration and backup restore
+preserve prompts, pairs, image links and aggregate counters; storage epochs keep
+stale queued writes from being replayed into restored data. There is no
+historical backfill from counters, scan dates, image timestamps or saved recipes.
+
+[Testing](../TESTING.md#durable-execution-ledger) describes the real CPU host
+matrix, historical failure proof, privacy checks and cleanup. Python deployment
+requires restarting ComfyUI.
+
+## What the existing UI proves
+
+`prompts.run_count` and `prompt_pairs.run_count` remain aggregate usage counters,
+not completed-run counts. `last_used_at` is the latest aggregate update; legacy
+migration may have estimated counts from linked images. `generated_images`
+identifies linked files; image counts and insertion times are not run counts or
+execution times. Image/pair association creation time can reflect an old-image
+scan. A saved pair's creation time is its recipe date.
+
+The native Prompt Library shows saved negative variants, linked images and
+aggregate uses. It does not expose the ledger as chronological dots or repeat-run
+branches. `PromptTracker` identities and PNG recipe evidence remain separate
+from durable host job identity.
+
+## Immutable contexts and output references
+
+Prompt contexts retain bounded raw/effective positive, negative and composition
+snapshots with nullable recipe navigation hints and explicit evidence provenance.
+Unknown text remains SQL NULL; a known empty negative remains empty text. Cached
+evidence belongs to its exact cache provider and attempt, rather than inheriting
+unrelated recent prompts. Deletion and restore can remove navigation hints without
+rewriting historical snapshots or rebinding them to current library rows.
+
+Schema v3 adds `execution_output_capture` and `execution_run_outputs` without
+backfilling older runs or modifying gallery image links. At final queue completion,
+the writer attaches bounded output evidence from that attempt's final host history
+to its immutable run ID. The internal `get_execution_outputs(run_id)` reader
+returns state, reason, format version, source and ordered references. A complete
+capture with `none_reported` means the final host history reported no images;
+`not_captured`, partial, unavailable and conflicting evidence remain explicit.
+Failed and interrupted jobs can still have complete evidence for images reported
+before termination. Running jobs do not gain a final capture merely because an
+image node finished.
+
+Each reference records output node ID, image ordinal, folder type, subfolder,
+filename, path style and an opaque configured-root key. A cached repeat can retain
+the same locators under a new run. Locators do not verify file bytes, continued
+existence, a newly generated file or membership in any prompt context: the reader
+reports file identity as `unverified`, generation and context membership as
+`unknown`, and temp outputs as transient. Capture never reads, resolves or serves
+the files and adds no public route. PNG metadata is not required. Same-path file
+replacement and existing gallery provenance retain their separate semantics.
+
+## Remaining history work
+
+Define a deduplicated run/gallery-image relationship that preserves existing image
+provenance and same-path replacement protection. Add a chronological timeline only
+after its navigation and evidence semantics can honestly represent prompt-level
+execution chronology, batched outputs and repeat-run branches.

docs/prompt-inspector.md

@@ -0,0 +1,40 @@
+# Prompt inspection and sending
+
+The opt-in workspace at `/prompt_manager/workspace#/images` shows full positive
+and negative prompts in both Image details and the full-image preview. Existing
+gallery and metadata tools remain available during the application rebuild.
+
+Each prompt shows its evidence source: recorded execution, workflow inference,
+or embedded generation parameters. Path-only saved-library associations are
+marked unverified and cannot be sent: a filename alone cannot prove that an image
+has not been replaced. Missing, ambiguous, and unreadable prompts remain visible
+as distinct states. An explicitly
+empty negative is a resolved value and can clear a destination's negative field.
+
+Use the blue paper-plane action beside a prompt, or **Send prompt pair**. Keep a
+ComfyUI tab open under the same origin and deployment prefix. Choose a window and
+compatible Prompt Manager node, select **Preview changes**, review the complete
+before/after text and preserved prepend/append composition, then **Apply to
+ComfyUI**. The dialog reports success only after a matching acknowledgement.
+It does not load workflows or queue generation. If confirmation is lost, inspect
+ComfyUI and preview again before another attempt. Closing the dialog cannot undo
+an update already sent.
+
+## Read-only API
+
+`GET /prompt_manager/images/inspect/{relative_path}?root={root_index}` requires
+an explicit configured root, including when two roots contain the same filename.
+The response contains `image` identity and `prompts.positive` / `prompts.negative`.
+Each role has `status` (`resolved`, `absent`, `unresolved`, or `ambiguous`), `text`
+(string only when resolved, otherwise null), and `source`. Inspection does not
+index files, alter saved prompts, run scans, or decode image pixels. PNG text and
+video probe output are bounded; unsupported or malformed evidence remains
+unresolved rather than becoming sendable placeholder text. Videos are read from
+one container carrier: the per-key tags written by core SaveVideo/SaveWEBM and
+current VideoHelperSuite (`prompt`, `workflow`, `parameters`,
+`comfyassets_prompt_pairs`), else the old VideoHelperSuite `comment` envelope,
+else `description`. The admin scan reads the same carrier, so Images and
+ingestion agree on a video's prompts (see TESTING.md for the order and budgets;
+the live monitor processes images only).
+
+Python route changes require restarting ComfyUI. Browser assets require a refresh.

docs/prompt-send-bridge.md

@@ -0,0 +1,49 @@
+# Prompt sending bridge
+
+Issue #202 establishes the transport and host adapter before the redesigned
+inspector displays prompt text. Gallery paper-plane controls and normalized
+metadata follow in a separate slice; existing node layouts are unchanged.
+
+## Shared transport
+
+PromptManager consumes the exact Toolkit 0.2.0 artifact recorded in
+`web/js/shared/toolkit/provenance.json`. The shared module has no ComfyUI imports.
+Its client discovers explicit host sessions and compatible destinations, requests
+a preview, and applies that preview using a one-use, client-bound, expiring ticket.
+Success requires an acknowledgement with verified destination values.
+
+BroadcastChannel connects independently opened tabs on the same origin and storage
+partition. Deployment prefixes have separate channel namespaces. Exact origin,
+protocol and peer checks reject unrelated messages. Same-origin scripts are
+trusted; this transport is not authentication against other scripts on that origin.
+Unsupported messaging is reported without breaking existing ComfyUI nodes.
+
+An apply timeout or abort after posting means the outcome is unknown. It cannot
+prove the host cancelled the write. The sender must inspect a fresh destination
+preview before deciding to send again; automatic retry must not overwrite edits.
+
+## PromptManager adapter
+
+Classic, Text, Focused and Split accept positive text; +/- accepts either role or
+both. Plain CLIP encoders are not assigned roles by title or position. An omitted
+role remains untouched; an explicitly empty string clears only the selected
+field. Unknown/ambiguous metadata must not become an empty string.
+
+Previews show original and requested text plus positive prepend/append composition.
+Connected composition is marked unknown rather than guessing upstream values.
+Connected destination fields reject writes. Applying rechecks graph, node, widgets,
+values and connections. Ordinary serialization cache synchronization is accepted
+only when it preserves live values. Both requested roles are checked before writing.
+
+Widget values, visible inputs, serialization and composed previews are synchronized.
+Callback failures restore local fields; arbitrary external callback side effects
+cannot be undone. Sending never queues execution or loads/replaces a workflow.
+
+## Lifecycle and validation
+
+Page hiding closes the host and invalidates tickets; restored pages mount one fresh
+host. Client disposal clears pending timers/listeners and closes its channel.
+Shared protocol contracts run against the actual bundled file, alongside adapter
+units, cross-browser tab workflows and real Classic/Nodes 2.0 scenarios. Pinned
+PromptManager-only, KikoTools-only and combined disposable hosts verify coexistence
+and workflow reload without modifying either live installation.

docs/reusable-prompt-pairs.md

@@ -0,0 +1,111 @@
+# Reusable positive and negative prompts
+
+Tracking: [#193](https://github.com/ComfyAssets/ComfyUI_PromptManager/issues/193).
+This work builds on the merged Split node in PR #192. Keep the feature unmerged
+until maintainer live testing is complete.
+
+## Storage contract
+
+Keep existing prompt IDs, text hashes, metadata, image links and node schemas.
+A component can serve either role. Ordered combinations refer to shared prompt
+rows; an explicitly empty negative uses a null reference. Missing or ambiguous
+metadata does not mean an empty negative. Pair metadata belongs to the
+combination; saving another combination does not overwrite component metadata.
+Existing trim/lowercase hash normalization remains unchanged.
+
+Saved components are immutable while referenced. To change text, create or reuse
+a component and save a new combination. Archive combinations that have generation
+history. Pruning and legacy editing/deletion must respect these dependencies.
+Historical workflow snapshots remain intact, including their original text.
+
+## Migration and recovery
+
+Create pair tables after existing legacy rebuild migrations, in an atomic,
+repeatable migration. Never scan media during schema initialization. Test fresh,
+legacy and current databases, repeated startup, injected failures and retries.
+Preserve old backup compatibility. Migrate and validate a disposable backup copy
+before replacing live database pages. Reject broken pair schemas, foreign keys,
+triggers and views. Use SQLite backup APIs so committed WAL data is included.
+
+## Metadata and scanning
+
+Resolve positive and negative from the same sampler/output context. Track both
+node ID and output slot; bound graph traversal and text construction. Unknown
+conditioning operations and conflicting sampler candidates remain unresolved.
+Never guess that the first encoder or most recent negative belongs to an image.
+
+Import high-confidence pairs atomically with their image association and source
+provenance. Repeated scans must not duplicate components, pairs or image links,
+regenerate thumbnails, or increment generation run counts. Preserve existing
+conflicting/manual associations for review. Report new, enriched, unchanged,
+ambiguous, unresolved and failed results separately.
+
+## Live node
+
+Use the shared toolkit and Split foundation. Positive and negative editors have
+equal size. Only positive receives prepend, append and automatic LoRA triggers.
+Keep upstream connections authoritative. Library selection previews both fields
+before an explicit load. Saving text changes creates another reusable combination.
+Verify Classic and Nodes 2.0 startup, switching both directions, reload, caching,
+linked inputs, popup behavior and lifecycle cleanup before maintainer testing.
+
+## Delivery gate
+
+Keep all existing coverage floors and test suites. Record regression failures
+before implementing each behavior. Tests use disposable state and the managed
+CPU host; never mutate the maintainer's running ComfyUI or personal database.
+Python changes require a ComfyUI restart. Live approval is required before merge.
+
+## Live testing
+
+1. Download a database backup from Admin before the first restart. The additive
+   schema migration runs on startup; existing workflows and prompt IDs stay valid.
+2. Restart ComfyUI, reload the browser, and add **Prompt Manager (+/-)** from
+   **ComfyAssets / Prompts**. Connect CLIP and the two conditioning outputs to the
+   sampler's positive and negative inputs.
+3. Enter both texts, save a pair, then keep positive unchanged and change negative.
+   Recent should show both combinations with distinct negative excerpts. Selecting
+   one previews both; **Use pair** loads both together.
+4. Test category/tag autocomplete and metadata opt-in, resizing, light/dark,
+   switching Classic / Nodes 2.0 both ways, and workflow reload. Connected fields
+   stay authoritative; run the node to save connected metadata values.
+5. Queue the same recipe twice and change only negative once. Usage should count
+   dequeued jobs even when ComfyUI caches the node. Prepend, append and LoRA triggers
+   affect positive only. Gallery text is labelled as recorded during execution or
+   recovered from workflow metadata.
+6. Run an output metadata scan to enrich existing images. Repeat it: pairs and
+   links must remain deduplicated, with unchanged results rather than new links.
+   Conflicting or ambiguous files need review and are preserved.
+
+Fresh executions embed bounded text evidence in PNG metadata. A cached node may
+not emit fresh evidence, so its graph-derived text is explicitly marked inferred.
+Multiple distinct batch records without an image anchor remain unresolved.
+Rescanning after interruption is safe through idempotent per-file writes; this
+version does not persist a filesystem scan cursor across server restarts.
+The API can request cancellation with `POST /prompt_manager/scan/cancel`; the
+existing dashboard Hide button continues to hide progress without cancelling.
+
+Pair APIs are under `/prompt_manager/pairs`: POST saves a combination, GET lists
+and filters, GET/PATCH/DELETE `/{id}` reads, edits metadata or removes an unused
+pair. Changing text creates another combination. Null metadata updates preserve
+existing values; empty strings/lists clear supported fields. Archive a pair with
+`PATCH {"archived":true}`. GET `/export` provides versioned JSON containing both
+texts and pair metadata; this is an export format, not a JSON import endpoint.
+
+## Gallery redesign concept retained
+
+The maintainer approved the interactive
+[negative variant timeline concept](concepts/prompt-variant-timeline.html) on
+2026-10-17. Preserve it for the later gallery redesign; it is not production UI
+and does not expand the current node work.
+
+Keep one positive parent and one shared gallery, with red branches for reusable
+negative variants and smaller dots for repeat runs. Show the selected negative
+in full directly beneath the unchanged positive; selecting another variant or
+run updates that text and filters the shared gallery. Comparison shows both
+negatives. Distinguish explicitly empty negatives from an all-variants view and
+saved variants from executed runs. Production history needs persisted execution
+records; the prototype's timeline uses illustrative sample data.
+
+The Admin upgrade/scan notice remains a separate follow-up: explain automatic
+schema migration and offer an optional metadata enrichment scan or Later.

docs/screenshots/208/README.md

@@ -0,0 +1,13 @@
+# Native Library visual evidence
+
+Captured from the production workspace using disposable synthetic browser-test
+fixtures. No personal images, library, or running ComfyUI state was accessed.
+Solid-color media are intentional test images.
+
+- `library-dark.png`: grouped positive shelf, dark appearance.
+- `library-light.png`: the same shelf, light appearance.
+- `history-selected-negative.png`: selected negative directly below its positive,
+  saved-variant branches and the filtered shared gallery.
+- `history-mobile.png`: the same history at a 390-pixel viewport.
+
+These screenshots document implemented behavior, rather than a design mockup.

prompt_manager.py

@@ -6,6 +6,11 @@ with persistent prompt storage and search capabilities.
 import time
 from typing import Any, Dict, Optional, Tuple
 
+if __package__:
+    from .utils.execution_context_capture import record_runtime_context
+else:
+    from utils.execution_context_capture import record_runtime_context
+
 try:
     from comfy.comfy_types import IO, ComfyNodeABC, InputTypeDict
 except ImportError:
@@ -256,7 +261,11 @@ class PromptManager(PromptManagerBase, ComfyNodeABC):
         )
 
         self.logger.info(f"CLIP encoding completed, text: {repr(encoding_text)[:80]}")
-        return (conditioning, encoding_text)
+        result = (conditioning, encoding_text)
+        record_runtime_context(
+            type(self).__name__, text, prepend_text, append_text, encoding_text
+        )
+        return result
 
     @classmethod
     def IS_CHANGED(
@@ -296,3 +305,7 @@ class PromptManagerFocused(PromptManager):
                 {**options, "default": "", "forceInput": True},
             )
         return inputs
+
+
+class PromptManagerSplit(PromptManagerFocused):
+    """The persistent library layout shares Focused encoding and storage."""