Sync Space from GitHub 698a7ac548e2c8414525dfb0c58acd8c6755a0bd

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

@@ -29,6 +29,7 @@ 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
@@ -36,6 +37,7 @@ def test_tier1_benchmark_workflow_contract():
     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"):