fix(eval): preserve incomplete observation evidence

benchmarks/schemas/observation-evidence.schema.json

@@ -32,10 +32,10 @@
     "dataset": {"type": "string", "minLength": 1},
     "condition": {"type": "string", "minLength": 1},
     "snapshot_id": {
-      "type": "string",
+      "type": ["string", "null"],
       "pattern": "^sha256:[0-9a-f]{64}$"
     },
-    "harbor_version": {"type": "string", "minLength": 1},
+    "harbor_version": {"type": ["string", "null"], "minLength": 1},
     "eval_args": {"$ref": "#/$defs/eval_args"},
     "job": {"$ref": "#/$defs/job"},
     "runtime_snapshot": {"$ref": "#/$defs/runtime_snapshot"},
@@ -55,6 +55,11 @@
       },
       "then": {
         "properties": {
+          "snapshot_id": {
+            "type": "string",
+            "pattern": "^sha256:[0-9a-f]{64}$"
+          },
+          "harbor_version": {"type": "string", "minLength": 1},
           "trials": {
             "type": "array",
             "items": {

benchmarks/validation/test_observation_artifact_binding.py

@@ -231,7 +231,7 @@ def test_observation_rejects_missing_snapshot_id(
         ],
     }
     job_path = _write_observation_job(
-        tmp_path, job, snapshot_id=None, harbor_version=_HARBOR_VERSION
+        tmp_path, job, snapshot_id=None, harbor_version=None
     )
     result_path = _write_result(tmp_path)
 
@@ -245,6 +245,22 @@ def test_observation_rejects_missing_snapshot_id(
 
     assert evidence["status"] == "INCOMPLETE"
     assert any("snapshot_id" in f and "missing" in f for f in failures)
+    assert any("harbor_version" in f and "missing" in f for f in failures)
+
+    schema = json.loads(
+        (
+            Path(__file__).resolve().parents[1]
+            / "schemas"
+            / "observation-evidence.schema.json"
+        ).read_text(encoding="utf-8")
+    )
+    from jsonschema import Draft202012Validator
+
+    assert list(Draft202012Validator(schema).iter_errors(evidence)) == []
+
+    evidence["status"] = "VALID"
+    errors = list(Draft202012Validator(schema).iter_errors(evidence))
+    assert errors, "VALID evidence must retain reproducibility bindings"
 
 
 def test_observation_rejects_mismatched_harbor_version(

tests/component/providers/polynomial/test_polynomial_map_inverse_synthesize.py

@@ -302,6 +302,7 @@ def test_ring_mismatches_fail_closed(polynomial_services) -> None:
         assert result.artifact_uris == (), mutation
 
 
+@pytest.mark.timeout(60)
 def test_corrupted_found_candidate_does_not_verify(
     authorized_polynomial_services,
 ) -> None: