주기 실회기 검증과 G7 종료계약 보강

scripts/test_g7_external_proof.py

@@ -44,6 +44,7 @@ def human_pack() -> dict[str, object]:
     )
     observations = []
     axes = ("goal", "task", "bond")
+    categories = ("low", "medium", "high")
     for session_index in range(50):
         participant = f"held-{session_index % 30:03d}"
         for axis_index, axis in enumerate(axes):
@@ -60,8 +61,16 @@ def human_pack() -> dict[str, object]:
                     "voice_enabled_status": "observed",
                     "voice_enabled_score": target,
                     "labels": [
-                        {"labeler_key": "labeler-001", "score": target},
-                        {"labeler_key": "labeler-002", "score": target},
+                        {
+                            "labeler_key": "labeler-001",
+                            "score": target,
+                            "category": categories[axis_index],
+                        },
+                        {
+                            "labeler_key": "labeler-002",
+                            "score": target,
+                            "category": categories[axis_index],
+                        },
                     ],
                 }
             )
@@ -108,6 +117,7 @@ def human_pack() -> dict[str, object]:
         "reliability": {
             "labeler_keys": ["labeler-001", "labeler-002"],
             "reported_icc": 1.0,
+            "reported_categorical_kappa": 1.0,
             "report_sha256": _sha(112),
         },
         "observations": observations,
@@ -432,6 +442,25 @@ class G7ExternalProofTests(unittest.TestCase):
         self.assertEqual([], errors)
         self.assertTrue(gain["passed"])
         self.assertGreaterEqual(gain["held_out_participants"], 30)
+        self.assertGreaterEqual(gain["recomputed_categorical_kappa"], 0.70)
+
+    def test_human_pack_cannot_bypass_the_required_categorical_kappa(self) -> None:
+        errors: list[str] = []
+        payload = human_pack()
+        reliability = payload["reliability"]
+        assert isinstance(reliability, dict)
+        reliability.pop("reported_categorical_kappa")
+        observations = payload["observations"]
+        assert isinstance(observations, list)
+        for observation in observations:
+            labels = observation["labels"]
+            for label in labels:
+                label.pop("category")
+
+        gain = self.checker.validate_human_gain(payload, errors)
+
+        self.assertEqual({}, gain)
+        self.assertIn("human_gain:invalid:ValidationError", errors)
 
     def test_windows_host_topology_passes_without_weakening_compose(self) -> None:
         errors: list[str] = []