feat(laboratory): compare M48S runtime hardening

tests/test_m48s_fixed_class_detector_lab.py

@@ -38,7 +38,7 @@ def test_m48s_lab_seals_visual_comparison_and_load_evidence(tmp_path: Path) -> N
     assert manifest["result_id"] == result.result_id
     assert result.result_id.endswith(identity_digest)
     assert manifest["identity_sha256"] == identity_digest
-    assert len(manifest["artifacts"]) == 31
+    assert len(manifest["artifacts"]) == 35
     assert manifest["method"]["completeness"] == "complete"
     assert manifest["bounded_question_accepted"] is True
     assert manifest["ground_truth"] is False
@@ -66,6 +66,20 @@ def test_m48s_lab_seals_visual_comparison_and_load_evidence(tmp_path: Path) -> N
     assert max(integrated["queue_high_watermarks"].values()) <= 2
     assert integrated["additional_inference_passes"] == 0
     assert integrated["failures"] == 0
+    hardening = manifest["metrics"]["runtime_hardening"]
+    assert hardening["schema_version"] == "missioncore.m48s-runtime-hardening-comparison/v1"
+    assert hardening["baseline"]["delivered_world_states"] == 4_480
+    assert hardening["baseline"]["superseded_frames"] == 9
+    assert hardening["hardened"]["delivered_world_states"] == 4_488
+    assert hardening["hardened"]["superseded_frames"] == 1
+    assert hardening["baseline"]["rolling_maximum_ms"] == 762.638263
+    assert hardening["hardened"]["rolling_maximum_ms"] == 28.044958
+    assert hardening["startup"]["baseline"]["detector_ms"] == 397.356888
+    assert hardening["startup"]["prewarmed"]["detector_ms"] == 27.662887
+    assert hardening["startup"]["baseline"]["world_state_ms"] == 449.893287
+    assert hardening["startup"]["prewarmed"]["world_state_ms"] == 60.516957
+    assert hardening["startup"]["prewarm_duration_ms"] == 420.721918
+    assert hardening["startup"]["validation_frames"] == 1_000
 
     catalog = json.loads((result.result_root / "catalog.json").read_text("utf-8"))
     assert catalog["frame_count"] == len(FRAME_IDS) == 11
@@ -85,7 +99,7 @@ def test_m48s_lab_seals_visual_comparison_and_load_evidence(tmp_path: Path) -> N
     )
     proof = verify_laboratory_evidence_result(definition, result.result_root)
     assert proof["result_id"] == result.result_id
-    assert proof["artifact_count"] == 31
+    assert proof["artifact_count"] == 35
 
     with pytest.raises(M48SFixedClassDetectorLabError, match="already exists"):
         build_m48s_fixed_class_detector_lab(

tests/test_m48s_fixed_class_detector_lab_api.py

@@ -47,20 +47,19 @@ def test_m48s_lab_api_projects_verified_result_frame_and_camera(tmp_path: Path)
         result.json()["metrics"]["integrated_world_state"]["world_state_completion_age_p95_ms"]
         == 74.733648
     )
+    hardening = result.json()["metrics"]["runtime_hardening"]
+    assert hardening["baseline"]["delivered_world_states"] == 4_480
+    assert hardening["hardened"]["delivered_world_states"] == 4_488
+    assert hardening["startup"]["prewarmed"]["world_state_ms"] == 60.516957
     assert result.json()["ground_truth"] is False
     assert len(result.json()["frames"]) == 11
 
-    timeline = client.get(
-        f"/api/v1/laboratory/m48s/fixed-class-detector/{result_id}/timeline"
-    )
+    timeline = client.get(f"/api/v1/laboratory/m48s/fixed-class-detector/{result_id}/timeline")
     assert timeline.status_code == 200
     assert timeline.json()["frame_count"] == 4_489
     assert timeline.json()["world_state_frame_count"] == 4_481
     assert timeline.json()["superseded_frame_count"] == 8
-    assert (
-        timeline.json()["camera_point_delivery"]
-        == "factory-kb4-causal-registered-accumulation"
-    )
+    assert timeline.json()["camera_point_delivery"] == "factory-kb4-causal-registered-accumulation"
     assert timeline.json()["camera_point_window_seconds"] == 2.0
     assert timeline.json()["camera_point_sample_limit"] == 20_000
     chunk = client.get(
@@ -72,9 +71,7 @@ def test_m48s_lab_api_projects_verified_result_frame_and_camera(tmp_path: Path)
     assert replay_frame["world_state_available"] is True
     assert replay_frame["camera_projection"] == "factory-kb4-exact"
     assert replay_frame["camera_projected_sample_count"] > 0
-    assert any(
-        item["semantic_hint"] == "dog" for item in replay_frame["camera_proposals"]
-    )
+    assert any(item["semantic_hint"] == "dog" for item in replay_frame["camera_proposals"])
     camera_points = client.get(
         f"/api/v1/laboratory/m48s/fixed-class-detector/{result_id}"
         "/timeline/frames/253/camera-points"
@@ -82,9 +79,7 @@ def test_m48s_lab_api_projects_verified_result_frame_and_camera(tmp_path: Path)
     assert camera_points.status_code == 200
     camera_point_payload = camera_points.json()
     assert camera_point_payload["schema_version"] == "missioncore.m48s-camera-point-overlay/v1"
-    assert camera_point_payload["projection"] == (
-        "factory-kb4-causal-registered-accumulation"
-    )
+    assert camera_point_payload["projection"] == ("factory-kb4-causal-registered-accumulation")
     assert camera_point_payload["source_frame_count"] > 1
     assert camera_point_payload["sample_count"] > replay_frame["camera_projected_sample_count"]
     assert camera_point_payload["sample_count"] <= 20_000