Fix benchmark manifest Git SHA

.github/workflows/tier1-benchmark.yml

@@ -223,7 +223,7 @@ jobs:
             --memory=4Gi \
             --cpu=2 \
             --service-account="$SERVICE_ACCOUNT" \
-            --set-env-vars="INTERACTION_DB_REPO=${INTERACTION_DB_REPO},INTERACTION_DB_TAG=${INTERACTION_DB_TAG},INTERACTION_DB_SHA256=${INTERACTION_DB_SHA256},BENCHMARK_FORCE_OS_EXIT=1" \
+            --set-env-vars="INTERACTION_DB_REPO=${INTERACTION_DB_REPO},INTERACTION_DB_TAG=${INTERACTION_DB_TAG},INTERACTION_DB_SHA256=${INTERACTION_DB_SHA256},BENCHMARK_FORCE_OS_EXIT=1,BENCHMARK_GIT_COMMIT=${GITHUB_SHA}" \
             --set-secrets="HF_TOKEN=HF_TOKEN:latest" \
             --args="$ARGS"
 
@@ -285,7 +285,7 @@ jobs:
               "run_id": manifest.get("run_id") == os.environ["RUN_ID"],
               "dataset_revision": manifest.get("dataset_revision") == os.environ["DATASET_REVISION"],
               "sample_size": 0 < sample_size <= expected_limit,
-              "git_commit": bool(manifest.get("git_commit")),
+              "git_commit": manifest.get("git_commit") not in ("", "0000000", None),
               "model_ids": bool(manifest.get("model_ids", {}).get("ner")),
               "ddinter_repo": manifest.get("ddinter_db", {}).get("repo") == os.environ["INTERACTION_DB_REPO"].strip("\r\n"),
               "ddinter_tag": manifest.get("ddinter_db", {}).get("tag") == os.environ["INTERACTION_DB_TAG"].strip("\r\n"),

eval/run_benchmark.py

@@ -787,6 +787,9 @@ def ddinter_metadata_from_args(args: argparse.Namespace) -> dict:
 
 
 def _git_commit() -> str:
+    env_commit = os.environ.get("BENCHMARK_GIT_COMMIT") or os.environ.get("GITHUB_SHA")
+    if env_commit:
+        return env_commit
     try:
         completed = subprocess.run(
             ["git", "rev-parse", "HEAD"],

tests/eval/test_run_benchmark.py

@@ -366,6 +366,13 @@ def test_manifest_includes_ddinter_release_metadata():
     assert manifest["concurrency"] == 8
 
 
+def test_git_commit_uses_benchmark_env_when_git_metadata_missing(monkeypatch):
+    monkeypatch.setenv("BENCHMARK_GIT_COMMIT", "e9f24645baf2d64a0504ee5b03c7af9f767b74cb")
+    monkeypatch.setattr(run_benchmark.subprocess, "run", lambda *args, **kwargs: (_ for _ in ()).throw(RuntimeError))
+
+    assert run_benchmark._git_commit() == "e9f24645baf2d64a0504ee5b03c7af9f767b74cb"
+
+
 @pytest.mark.asyncio
 async def test_ensure_ddinter_db_downloads_from_github_release(monkeypatch, tmp_path: Path):
     checks = iter([False, True])

tests/test_tier1_benchmark_workflow.py

@@ -29,13 +29,15 @@ def test_tier1_benchmark_workflow_contract():
     assert "INTERACTION_DB_TAG" in workflow
     assert "INTERACTION_DB_SHA256" in workflow
     assert "BENCHMARK_FORCE_OS_EXIT=1" in workflow
+    assert "BENCHMARK_GIT_COMMIT=${GITHUB_SHA}" in workflow
     assert 'tr -d "\\r\\n"' in workflow
     assert 'INTERACTION_DB_REPO="$(printf' in workflow
     assert 'INTERACTION_DB_TAG="$(printf' in workflow
     assert "--record-timeout-seconds" in workflow
     assert "hf://buckets/SPerva/pillchecker-experiments" in workflow
     assert 'bucket = os.environ["BENCHMARK_BUCKET"]' in workflow
     assert "0 < sample_size <= expected_limit" in workflow
+    assert 'manifest.get("git_commit") not in ("", "0000000", None)' in workflow
     assert "len(predictions) == sample_size" in workflow
 
     for artifact in ("manifest.json", "results.json", "predictions.jsonl", "errors.jsonl", "summary.md"):