diff --git a/fastvideo/tests/ssim/inference_similarity_utils.py b/fastvideo/tests/ssim/inference_similarity_utils.py
index a3e419d35..9ae9b0d46 100644
--- a/fastvideo/tests/ssim/inference_similarity_utils.py
+++ b/fastvideo/tests/ssim/inference_similarity_utils.py
@@ -9,8 +9,7 @@ from logging import Logger
 
 from fastvideo import VideoGenerator
 from fastvideo.tests.ssim.bootstrap_references import (
-    xfail_missing_reference_in_bootstrap_mode,
-)
+    xfail_missing_reference_in_bootstrap_mode, )
 from fastvideo.tests.ssim.reference_utils import (
     build_generated_output_dir,
     build_reference_folder_path,
@@ -26,6 +25,10 @@ DEVICE_MAPPINGS = (
     ("L40S", "L40S"),
     ("H100", "H100"),
     ("H200", "H200"),
+    # GB200 must precede B200: the lookup is a substring scan and
+    # "B200" matches "NVIDIA GB200", so the coarser pattern would
+    # otherwise claim every Grace-Blackwell host.
+    ("GB200", "GB200"),
     ("B200", "B200"),
 )
 
@@ -115,11 +118,9 @@ def _assert_similarity(
             reference_folder=reference_folder,
             artifact_kind=artifact_kind,
         )
-        error_msg = (
-            f"Reference video folder does not exist: {reference_folder}\n"
-            f"To download reference videos, run:\n"
-            f"  python fastvideo/tests/ssim/reference_videos_cli.py download"
-        )
+        error_msg = (f"Reference video folder does not exist: {reference_folder}\n"
+                     f"To download reference videos, run:\n"
+                     f"  python fastvideo/tests/ssim/reference_videos_cli.py download")
         raise FileNotFoundError(error_msg)
 
     try:
@@ -167,15 +168,11 @@ def _assert_similarity(
     if not success:
         logger.error("Failed to write SSIM results to file")
 
-    assert mean_ssim >= min_acceptable_ssim, (
-        f"SSIM value {mean_ssim} is below threshold {min_acceptable_ssim} "
-        f"for {model_id} with backend {attention_backend_name}"
-    )
+    assert mean_ssim >= min_acceptable_ssim, (f"SSIM value {mean_ssim} is below threshold {min_acceptable_ssim} "
+                                              f"for {model_id} with backend {attention_backend_name}")
 
 
-def build_init_kwargs(
-    base_params: dict[str, object],
-) -> dict[str, object]:
+def build_init_kwargs(base_params: dict[str, object], ) -> dict[str, object]:
     init_kwargs: dict[str, object] = {
         "num_gpus": base_params["num_gpus"],
         "sp_size": base_params.get("sp_size", 1),
@@ -193,18 +190,14 @@ def build_init_kwargs(
         init_kwargs["text_encoder_precisions"] = base_params["text-encoder-precision"]
     if base_params.get("ltx2_vae_tiling"):
         init_kwargs["ltx2_vae_tiling"] = True
-        init_kwargs["ltx2_vae_spatial_tile_size_in_pixels"] = base_params.get(
-            "ltx2_vae_spatial_tile_size_in_pixels", 512
-        )
+        init_kwargs["ltx2_vae_spatial_tile_size_in_pixels"] = base_params.get("ltx2_vae_spatial_tile_size_in_pixels",
+                                                                              512)
         init_kwargs["ltx2_vae_spatial_tile_overlap_in_pixels"] = base_params.get(
-            "ltx2_vae_spatial_tile_overlap_in_pixels", 64
-        )
-        init_kwargs["ltx2_vae_temporal_tile_size_in_frames"] = base_params.get(
-            "ltx2_vae_temporal_tile_size_in_frames", 64
-        )
+            "ltx2_vae_spatial_tile_overlap_in_pixels", 64)
+        init_kwargs["ltx2_vae_temporal_tile_size_in_frames"] = base_params.get("ltx2_vae_temporal_tile_size_in_frames",
+                                                                               64)
         init_kwargs["ltx2_vae_temporal_tile_overlap_in_frames"] = base_params.get(
-            "ltx2_vae_temporal_tile_overlap_in_frames", 24
-        )
+            "ltx2_vae_temporal_tile_overlap_in_frames", 24)
     return init_kwargs
 
 
diff --git a/fastvideo/tests/ssim/test_device_reference_folder.py b/fastvideo/tests/ssim/test_device_reference_folder.py
new file mode 100644
index 000000000..89e7c99a0
--- /dev/null
+++ b/fastvideo/tests/ssim/test_device_reference_folder.py
@@ -0,0 +1,48 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Device-name -> reference-folder resolution.
+
+The lookup is a substring scan over an ordered table, which makes it
+order-sensitive in a way that is easy to get wrong: ``"B200"`` is a substring of
+``"NVIDIA GB200"``, so a table listing B200 first silently routes every
+Grace-Blackwell host to the B200 references. That is not a cosmetic mismatch --
+whichever host seeds a reference defines the baseline every later run on the
+*other* host is scored against.
+"""
+from __future__ import annotations
+
+import pytest
+
+from fastvideo.tests.ssim.inference_similarity_utils import DEVICE_MAPPINGS
+from fastvideo.tests.ssim.reference_utils import resolve_device_reference_folder
+
+
+@pytest.mark.parametrize(
+    ("device_name", "expected"),
+    [
+        ("NVIDIA GB200", "GB200_reference_videos"),
+        ("NVIDIA B200", "B200_reference_videos"),
+        ("NVIDIA H200", "H200_reference_videos"),
+        ("NVIDIA H100 80GB HBM3", "H100_reference_videos"),
+        ("NVIDIA L40S", "L40S_reference_videos"),
+        ("NVIDIA A40", "A40_reference_videos"),
+    ],
+)
+def test_device_resolves_to_its_own_folder(device_name: str, expected: str) -> None:
+    assert resolve_device_reference_folder(DEVICE_MAPPINGS, device_name=device_name) == expected
+
+
+def test_gb200_is_not_captured_by_the_b200_pattern() -> None:
+    """The regression this file exists for.
+
+    Asserting the mapping order directly, so reordering the table fails here
+    rather than silently scoring GB200 runs against B200 references.
+    """
+    patterns = [pattern for pattern, _ in DEVICE_MAPPINGS]
+    assert "GB200" in patterns, "GB200 missing from DEVICE_MAPPINGS"
+    assert patterns.index("GB200") < patterns.index("B200"), (
+        "GB200 must precede B200: the lookup is a substring scan, so B200 "
+        "would otherwise match 'NVIDIA GB200' first")
+
+
+def test_unknown_device_returns_none() -> None:
+    assert resolve_device_reference_folder(DEVICE_MAPPINGS, device_name="NVIDIA RTX 5090") is None
diff --git a/fastvideo/tests/ssim/test_glm_image_similarity.py b/fastvideo/tests/ssim/test_glm_image_similarity.py
index 87f8ab357..e2e315755 100644
--- a/fastvideo/tests/ssim/test_glm_image_similarity.py
+++ b/fastvideo/tests/ssim/test_glm_image_similarity.py
@@ -10,8 +10,7 @@ import torch
 
 from fastvideo.logger import init_logger
 from fastvideo.tests.ssim.inference_similarity_utils import (
-    run_text_to_video_similarity_test,
-)
+    run_text_to_video_similarity_test, )
 from fastvideo.tests.ssim.reference_utils import (
     get_cuda_device_name,
     resolve_device_reference_folder,
@@ -22,9 +21,7 @@ logger = init_logger(__name__)
 REQUIRED_GPUS = 1
 
 REPO_ROOT = Path(__file__).resolve().parents[3]
-LOCAL_WEIGHTS_DIR = Path(
-    os.getenv("GLM_IMAGE_LOCAL_WEIGHTS_DIR",
-              REPO_ROOT / "official_weights" / "glm_image"))
+LOCAL_WEIGHTS_DIR = Path(os.getenv("GLM_IMAGE_LOCAL_WEIGHTS_DIR", REPO_ROOT / "official_weights" / "glm_image"))
 GLM_IMAGE_MODEL_PATH = os.getenv("GLM_IMAGE_MODEL_DIR", str(LOCAL_WEIGHTS_DIR))
 
 device_reference_folder = resolve_device_reference_folder(
@@ -33,6 +30,10 @@ device_reference_folder = resolve_device_reference_folder(
         ("L40S", "L40S"),
         ("H100", "H100"),
         ("H200", "H200"),
+        # GB200 must precede B200: the lookup is a substring scan and
+        # "B200" matches "NVIDIA GB200", so the coarser pattern would
+        # otherwise claim every Grace-Blackwell host.
+        ("GB200", "GB200"),
         ("B200", "B200"),
     ),
     device_name=get_cuda_device_name(),
@@ -87,9 +88,7 @@ GLM_IMAGE_FULL_QUALITY_MODEL_TO_PARAMS = {
 
 
 def _has_weights() -> bool:
-    required = ["transformer", "vae", "text_encoder",
-                "vision_language_encoder", "processor", "tokenizer",
-                "scheduler"]
+    required = ["transformer", "vae", "text_encoder", "vision_language_encoder", "processor", "tokenizer", "scheduler"]
     return all((LOCAL_WEIGHTS_DIR / r).exists() for r in required)
 
 
diff --git a/fastvideo/tests/ssim/test_zimage_similarity.py b/fastvideo/tests/ssim/test_zimage_similarity.py
index 2a96d7efe..541b6ff6b 100644
--- a/fastvideo/tests/ssim/test_zimage_similarity.py
+++ b/fastvideo/tests/ssim/test_zimage_similarity.py
@@ -10,8 +10,7 @@ import torch
 
 from fastvideo.logger import init_logger
 from fastvideo.tests.ssim.inference_similarity_utils import (
-    run_text_to_video_similarity_test,
-)
+    run_text_to_video_similarity_test, )
 from fastvideo.tests.ssim.reference_utils import (
     get_cuda_device_name,
     resolve_device_reference_folder,
@@ -44,6 +43,10 @@ device_reference_folder = resolve_device_reference_folder(
         ("L40S", "L40S"),
         ("H100", "H100"),
         ("H200", "H200"),
+        # GB200 must precede B200: the lookup is a substring scan and
+        # "B200" matches "NVIDIA GB200", so the coarser pattern would
+        # otherwise claim every Grace-Blackwell host.
+        ("GB200", "GB200"),
         ("B200", "B200"),
     ),
     device_name=get_cuda_device_name(),
@@ -67,10 +70,7 @@ ZIMAGE_MODEL_TO_PARAMS: dict[str, dict[str, object]] = {
     },
 }
 
-ZIMAGE_FULL_QUALITY_MODEL_TO_PARAMS = {
-    model_id: dict(params)
-    for model_id, params in ZIMAGE_MODEL_TO_PARAMS.items()
-}
+ZIMAGE_FULL_QUALITY_MODEL_TO_PARAMS = {model_id: dict(params) for model_id, params in ZIMAGE_MODEL_TO_PARAMS.items()}
 
 
 @pytest.mark.skipif(
@@ -87,10 +87,8 @@ def test_zimage_similarity(
 ) -> None:
     is_hf_repo = "/" in ZIMAGE_MODEL_PATH and not ZIMAGE_MODEL_PATH.startswith("/")
     if not is_hf_repo and not os.path.isdir(ZIMAGE_MODEL_PATH):
-        pytest.skip(
-            f"Z-Image weights not found at {ZIMAGE_MODEL_PATH} "
-            "(set ZIMAGE_MODEL_DIR to override)"
-        )
+        pytest.skip(f"Z-Image weights not found at {ZIMAGE_MODEL_PATH} "
+                    "(set ZIMAGE_MODEL_DIR to override)")
 
     run_text_to_video_similarity_test(
         logger=logger,