From 75ed93a1c4cece49fb3ae3889e9919545e24c348 Mon Sep 17 00:00:00 2001
From: dylan <dylan@bitmind.ai>
Date: Sun, 20 Sep 2026 22:25:46 -0500
Subject: [PATCH 1/2] Use current dataset labels when reading cached samples

---
 src/gasbench/dataset/download/cache_io.py |  2 ++
 src/gasbench/dataset/iterator.py          |  2 ++
 tests/unit/test_cached_sample_labels.py   | 40 +++++++++++++++++++++++
 3 files changed, 44 insertions(+)
 create mode 100644 tests/unit/test_cached_sample_labels.py

diff --git a/src/gasbench/dataset/download/cache_io.py b/src/gasbench/dataset/download/cache_io.py
index 148d215..81cb3dc 100644
--- a/src/gasbench/dataset/download/cache_io.py
+++ b/src/gasbench/dataset/download/cache_io.py
@@ -61,6 +61,8 @@ def _load_dataset_from_cache(dataset, cache_dir: str = "/.cache/gasbench"):
     for filename in sample_files:
         try:
             metadata = sample_metadata.get(filename, {})
+            # The current registry owns labels; caches may predate a taxonomy change.
+            metadata = {**metadata, "media_type": dataset.media_type}
             filepath = os.path.join(samples_dir, filename)
 
             if dataset.modality == "image":
diff --git a/src/gasbench/dataset/iterator.py b/src/gasbench/dataset/iterator.py
index 3f7b12f..3cd9658 100644
--- a/src/gasbench/dataset/iterator.py
+++ b/src/gasbench/dataset/iterator.py
@@ -618,6 +618,8 @@ def extract_index(filename):
             for filename in sample_items:
                 file_path = os.path.join(samples_dir, filename)
                 metadata = metadata_map.get(filename, {})
+                # The current registry owns labels; caches may predate a taxonomy change.
+                metadata = {**metadata, "media_type": self.config.media_type}
 
                 if self.config.modality == "image":
                     try:
diff --git a/tests/unit/test_cached_sample_labels.py b/tests/unit/test_cached_sample_labels.py
new file mode 100644
index 0000000..5182019
--- /dev/null
+++ b/tests/unit/test_cached_sample_labels.py
@@ -0,0 +1,40 @@
+import json
+
+import pytest
+
+from gasbench.constants import media_type_to_label
+from gasbench.dataset.config import BenchmarkDatasetConfig
+from gasbench.dataset.download.cache_io import _load_dataset_from_cache
+from gasbench.dataset.iterator import DatasetIterator
+
+
+@pytest.mark.parametrize("loader", ["iterator", "download"])
+@pytest.mark.parametrize("payload", ["bytes", "frames", "lazy"])
+def test_current_video_label_overrides_stale_cache(tmp_path, loader, payload):
+    config = BenchmarkDatasetConfig(
+        name="example", path="owner/repo", modality="video", media_type="real",
+        source_format="mp4", media_per_archive=-1, archives_per_dataset=-1,
+    )
+    directory = tmp_path / "datasets" / config.name
+    samples = directory / "samples"
+    samples.mkdir(parents=True)
+    filename = "clip" if payload == "frames" else "clip.mp4"
+    if payload == "frames":
+        (samples / filename).mkdir()
+        (samples / filename / "frame.jpg").write_bytes(b"frame")
+    else:
+        (samples / filename).write_bytes(b"video")
+    (directory / "sample_metadata.json").write_text(json.dumps({filename: {
+        "media_type": "rendered", "source_file": "original.mp4",
+    }}))
+    (directory / "dataset_info.json").write_text("{}")
+    if loader == "iterator":
+        iterator = DatasetIterator(config, cache_dir=str(tmp_path), download=False,
+                                   lazy_read=payload == "lazy")
+        result = list(iterator)
+    else:
+        result = list(_load_dataset_from_cache(config, str(tmp_path)))
+    assert len(result) == 1
+    assert media_type_to_label(result[0]["media_type"], "video") == 0
+    assert result[0]["source_file"] == "original.mp4"
+    assert json.loads((directory / "sample_metadata.json").read_text())[filename]["media_type"] == "rendered"

From 431ab596e45c5c3bb3601be993d48cf3cfa292f7 Mon Sep 17 00:00:00 2001
From: dylan <dylan@bitmind.ai>
Date: Mon, 21 Sep 2026 10:58:18 -0500
Subject: [PATCH 2/2] Bump patch version to 0.9.10

---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index e64b5d6..7ba02d0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gasbench"
-version = "0.9.9"
+version = "0.9.10"
 description = "GASBench - ML model benchmark evaluation package"
 readme = "README.md"
 requires-python = ">=3.10"