diff --git a/tests/browser/full-inspector.spec.mjs b/tests/browser/full-inspector.spec.mjs
index 5909afd..421e291 100644
--- a/tests/browser/full-inspector.spec.mjs
+++ b/tests/browser/full-inspector.spec.mjs
@@ -26,9 +26,9 @@ test('inspection facts survive actual API to browser transport', async ({ page,
     const complete = result.complete;
     expect(Object.values(complete.generation).map(fact => fact.value)).toEqual(['9007199254740993', '28', '6.5', 'dpmpp_2m', 'karras', 'fixture-model.safetensors']);
     expect(Object.values(complete.generation).every(fact => fact.source === 'api-graph' && fact.evidence === 'configuration')).toBe(true);
-    // Ambiguous branches merge per field: the differing seed is ambiguous, the shared fields known.
+    // Differing branches show what the legacy parser shows: the first sampler's seed.
     expect(Object.values(result.ambiguous.generation).map(fact => [fact.status, fact.value])).toEqual([
-        ['ambiguous', null], ['known', '28'], ['known', '6.5'], ['known', 'dpmpp_2m'], ['known', 'karras'], ['known', 'fixture-model.safetensors']]);
+        ['known', '9007199254740993'], ['known', '28'], ['known', '6.5'], ['known', 'dpmpp_2m'], ['known', 'karras'], ['known', 'fixture-model.safetensors']]);
     expect(Object.values(result.ambiguous.generation).every(fact => fact.sampler_id === null)).toBe(true);
     expect(result['workflow-only'].generation.seed).toMatchObject({ status: 'known', value: '9007199254740993', source: 'workflow', evidence: 'configuration' });
     expect(result['workflow-only'].prompts.positive.source).toBe('saved-association');
diff --git a/tests/js/workspace-inspection-state.test.js b/tests/js/workspace-inspection-state.test.js
index fcb8e4d..bb884be 100644
--- a/tests/js/workspace-inspection-state.test.js
+++ b/tests/js/workspace-inspection-state.test.js
@@ -464,3 +464,57 @@ test('rooted controller keeps legacy descriptors with ignored noncloneable field
     let pending;assert.doesNotThrow(()=>{pending=controller.select(selected);});
     requests[0].resolve(envelope());await pending;assert.equal(controller.state.record.identity.key,'0:folder/a #%.png');controller.dispose();
 });
+
+// The maintainer's standard pipeline: one PromptManager pair feeds a base pass, an
+// upscale pass and a FaceDetailer, saved by a custom saver. Text and seeds are invented.
+function detailerPipelineGraph() {
+    return {
+        1: { class_type: 'CheckpointLoaderSimple', inputs: { ckpt_name: 'Base/example_model.safetensors' } },
+        5: { class_type: 'KSampler', inputs: { seed: 111, steps: 4, cfg: 1.0, sampler_name: 'lcm', scheduler: 'normal', denoise: 1.0, model: ['47', 0], positive: ['25', 0], negative: ['26', 0], latent_image: ['20', 0] } },
+        6: { class_type: 'VAEDecode', inputs: { samples: ['5', 0], vae: ['1', 2] } },
+        7: { class_type: 'PreviewImage', inputs: { images: ['6', 0] } },
+        8: { class_type: 'UpscaleModelLoader', inputs: { model_name: 'upscaler.pth' } },
+        9: { class_type: 'ImageUpscaleWithModel', inputs: { upscale_model: ['8', 0], image: ['6', 0] } },
+        10: { class_type: 'ImageScaleBy', inputs: { upscale_method: 'lanczos', scale_by: 0.75, image: ['9', 0] } },
+        11: { class_type: 'VAEEncode', inputs: { pixels: ['10', 0], vae: ['1', 2] } },
+        12: { class_type: 'KSampler', inputs: { seed: 222, steps: 6, cfg: 1.0, sampler_name: 'lcm', scheduler: 'normal', denoise: 0.45, model: ['47', 0], positive: ['25', 0], negative: ['26', 0], latent_image: ['11', 0] } },
+        13: { class_type: 'VAEDecode', inputs: { samples: ['12', 0], vae: ['1', 2] } },
+        17: { class_type: 'FaceDetailer', inputs: { seed: 333, steps: 4, cfg: 1.0, sampler_name: 'lcm', scheduler: 'normal', denoise: 0.4, image: ['13', 0], model: ['47', 0], clip: ['47', 1], vae: ['1', 2], positive: ['25', 0], negative: ['26', 0] } },
+        20: { class_type: 'EmptyLatentBatch', inputs: { width: 640, height: 1536, batch_size: 1 } },
+        25: { class_type: 'PromptManager', inputs: { text: 'a lighthouse at dusk, oil painting', category: '', tags: '', search_text: '', prepend_text: '', append_text: '', clip: ['47', 1] } },
+        26: { class_type: 'PromptManager', inputs: { text: 'blurry, watermark', category: '', tags: '', search_text: '', prepend_text: '', append_text: '', clip: ['47', 1] } },
+        47: { class_type: 'Lora Loader (LoraManager)', inputs: { text: '', model: ['1', 0], clip: ['1', 1] } },
+        48: { class_type: 'KikoSaveImage', inputs: { filename_prefix: ['49', 0], format: 'PNG', images: ['17', 0] } },
+        49: { class_type: 'KikoProjectFilePath', inputs: { project_name: 'example', filename: 'image' } },
+    };
+}
+
+test('workflow prompts and settings match the legacy parser when the server cannot decide', t => {
+    const parser = require('../../web/js/comfy-metadata.js'), { extractGenerationParams } = parser;
+    globalThis.ComfyMetadata = parser;
+    t.after(() => { delete globalThis.ComfyMetadata; });
+    const entry = image(), data = envelope(entry);
+    data.prompts = { positive: { status: 'ambiguous', text: null, source: null }, negative: { status: 'ambiguous', text: null, source: null } };
+    data.metadata.prompt_graph = { status: 'parsed', value: detailerPipelineGraph() };
+    data.resolution = { state: 'ambiguous', sampler_id: null, reason: 'Multiple sampler branches' };
+    const record = contract('toInspectionRecord')(data, entry, extractGenerationFacts);
+    const legacy = extractGenerationParams({ prompt: detailerPipelineGraph() });
+    assert.deepEqual(record.prompts.positive, { status: 'resolved', text: legacy.positivePrompt, source: 'graph-inferred' });
+    assert.deepEqual(record.prompts.negative, { status: 'resolved', text: legacy.negativePrompt, source: 'graph-inferred' });
+    assert.equal(record.prompts.positive.text, 'a lighthouse at dusk, oil painting');
+    assert.equal(record.prompts.negative.text, 'blurry, watermark');
+    for (const [field, key] of [['seed', 'seed'], ['steps', 'steps'], ['cfg', 'cfgScale'], ['sampler', 'sampler'], ['model', 'checkpoint']]) {
+        assert.equal(record.generation[field].status, 'known', field);
+        assert.equal(record.generation[field].value, String(legacy[key]), field);
+    }
+});
+
+test('recorded and saved prompts are never replaced by the workflow parse', t => {
+    globalThis.ComfyMetadata = require('../../web/js/comfy-metadata.js');
+    t.after(() => { delete globalThis.ComfyMetadata; });
+    const entry = image(), data = envelope(entry);
+    data.metadata.prompt_graph = { status: 'parsed', value: detailerPipelineGraph() };
+    const record = contract('toInspectionRecord')(data, entry, extractGenerationFacts);
+    assert.deepEqual(record.prompts.positive, { status: 'resolved', text: 'exact\npositive', source: 'execution' });
+    assert.equal(record.prompts.negative.text, 'blurry, watermark');
+});
diff --git a/web/js/workspace/inspection-state.mjs b/web/js/workspace/inspection-state.mjs
index 9b7fb6b..638cd8a 100644
--- a/web/js/workspace/inspection-state.mjs
+++ b/web/js/workspace/inspection-state.mjs
@@ -163,6 +163,36 @@ function libraryFor(value) {
     return library;
 }
 
+// The legacy workflow parse (comfy-metadata.js, as of #170) also feeds the gallery and
+// admin pages; it fills any prompt or setting the server could not decide, so this page
+// never shows less than they do. Recorded and saved prompts are never replaced.
+const KEPT_PROMPT_SOURCES = ['execution', 'library', 'saved-association'];
+const LEGACY_FACTS = [['seed', 'seed'], ['steps', 'steps'], ['cfg', 'cfgScale'], ['sampler', 'sampler'], ['model', 'checkpoint']];
+
+function comfyMetadata() {
+    return globalThis.ComfyMetadata || globalThis.window?.ComfyMetadata;
+}
+
+function withWorkflowParams(record, parser = comfyMetadata()) {
+    const graph = record.metadata.prompt_graph.status === 'parsed' ? record.metadata.prompt_graph.value : null;
+    const workflow = record.metadata.workflow_graph.status === 'parsed' ? record.metadata.workflow_graph.value : null;
+    if (typeof parser?.extractGenerationParams !== 'function' || (!graph && !workflow)) return;
+    const params = parser.extractGenerationParams({ prompt: graph, workflow });
+    const found = key => typeof params[key] === 'string' || typeof params[key] === 'number' ?
+        String(params[key]) !== '' && params[key] !== parser.PLACEHOLDERS?.[key] : false;
+    const prompts = { ...record.prompts };
+    for (const [role, key] of [['positive', 'positivePrompt'], ['negative', 'negativePrompt']]) {
+        const current = prompts[role];
+        if (!found(key) || (current?.status === 'resolved' && KEPT_PROMPT_SOURCES.includes(current.source))) continue;
+        prompts[role] = { status: 'resolved', text: params[key], source: 'graph-inferred' };
+    }
+    record.prompts = prompts;
+    for (const [field, key] of LEGACY_FACTS) {
+        if (!found(key)) continue;
+        record.generation[field] = { status: 'known', value: String(params[key]), source: graph ? 'api-graph' : 'workflow', evidence: 'configuration', reason: null, sampler_id: null };
+    }
+}
+
 function unavailableFacts() {
     return Object.fromEntries(FACT_FIELDS.map(key => [key, { status: 'unresolved', value: null, source: null, evidence: null, reason: null, sampler_id: null }]));
 }
@@ -194,6 +224,7 @@ export function projectInspectionPayload(result, identity, download, extractFact
     if (typeof extractFacts !== 'function') throw new TypeError('Inspection facts parser unavailable');
     record.generation = extractFacts({ prompt_graph: record.metadata.prompt_graph, workflow_graph: record.metadata.workflow_graph,
         parameters: record.metadata.parameters, recordFormats: record.metadata.record_formats, metadataStatus: record.metadata.status, resolution });
+    withWorkflowParams(record);
     if (record.library.status === 'linked' && record.library.positive_id !== null) record.historyTarget = `#/library/${record.library.positive_id}`;
     return record;
 }