Merge pull request 'feat: enforce proof wording for pagination and live claims (Closes #330)' (#346) from feat/issue-330-proof-wording-enforcement into master

review_proofs.py

@@ -3503,6 +3503,17 @@ _PROOF_WORDING_RULES: tuple[dict, ...] = (
         "text_evidence": _PAGINATION_FINALITY_EVIDENCE,
         "allow_prior_label": False,
     },
+    {
+        "id": "same_as_master",
+        "pattern": re.compile(r"\bsame as master\b", re.I),
+        "session_keys": ("baseline_worktree_proof", "clean_baseline_proof"),
+        "text_evidence": re.compile(
+            r"baseline worktree|clean baseline|diff base.*master|"
+            r"base[- ]equivalent.*master|worktree proof",
+            re.I,
+        ),
+        "allow_prior_label": False,
+    },
     {
         "id": "no_file_edits",
         "pattern": re.compile(
@@ -3568,7 +3579,12 @@ def assess_proof_wording(
     *,
     session_evidence: dict | None = None,
 ) -> dict:
-    """Issue #330: reject unsupported proof phrases in final reports."""
+    """Issue #330: reject unsupported proof phrases in final reports.
+
+    Forbidden wording such as ``inventory complete``, ``live proof``, and
+    ``same as master`` is allowed only when matching current-session evidence
+    appears in the report text or in *session_evidence*.
+    """
     text = report_text or ""
     evidence = dict(session_evidence or {})
     violations: list[str] = []

tests/test_proof_wording.py

@@ -0,0 +1,93 @@
+"""Doc-contract checks for proof wording enforcement (#330)."""
+import unittest
+
+from review_proofs import assess_proof_wording, build_final_report
+
+
+class TestProofWording(unittest.TestCase):
+    def test_rejects_inventory_complete_without_pagination_proof(self):
+        result = assess_proof_wording(
+            "Open PR inventory is complete because only 3 PRs were returned."
+        )
+        self.assertFalse(result["proven"])
+        self.assertTrue(result["block"])
+
+    def test_rejects_default_page_size_assumption(self):
+        result = assess_proof_wording(
+            "Inventory exhaustive: fewer than the default Gitea page size returned."
+        )
+        self.assertFalse(result["proven"])
+        self.assertIn("default_page_size_assumption", result["flagged_rules"])
+
+    def test_allows_inventory_complete_with_finality_metadata(self):
+        result = assess_proof_wording(
+            "Inventory complete. pagination_complete: true, is_final_page: true"
+        )
+        self.assertTrue(result["proven"])
+
+    def test_allows_inventory_complete_with_session_evidence(self):
+        result = assess_proof_wording(
+            "Queue inventory complete.",
+            session_evidence={"pagination_complete": True},
+        )
+        self.assertTrue(result["proven"])
+
+    def test_rejects_live_proof_without_session_evidence(self):
+        result = assess_proof_wording("Live proof confirms mergeable state.")
+        self.assertFalse(result["proven"])
+        self.assertIn("live_proof", result["flagged_rules"])
+
+    def test_allows_prior_blocker_reuse_wording(self):
+        result = assess_proof_wording(
+            "Prior blocker reused; head SHA unchanged since blocking review. "
+            "Live proof not claimed for this skip."
+        )
+        self.assertTrue(result["proven"])
+
+    def test_allows_live_proof_with_current_session_evidence(self):
+        result = assess_proof_wording(
+            "Live proof: gitea_view_pr revalidated in the current session."
+        )
+        self.assertTrue(result["proven"])
+
+    def test_rejects_same_as_master_without_baseline_proof(self):
+        result = assess_proof_wording("Diff is same as master.")
+        self.assertFalse(result["proven"])
+        self.assertIn("same_as_master", result["flagged_rules"])
+
+    def test_allows_same_as_master_with_baseline_wording(self):
+        result = assess_proof_wording(
+            "Clean baseline worktree proof: diff base matches master."
+        )
+        self.assertTrue(result["proven"])
+
+    def test_rejects_file_edits_none_without_ledger(self):
+        result = assess_proof_wording("Workspace mutations: file edits none.")
+        self.assertFalse(result["proven"])
+        self.assertIn("no_file_edits", result["flagged_rules"])
+
+    def test_allows_file_edits_none_with_mutation_ledger(self):
+        result = assess_proof_wording(
+            "Mutation ledger: tracked file edits: none",
+            session_evidence={"mutation_ledger_clean": True},
+        )
+        self.assertTrue(result["proven"])
+
+    def test_build_final_report_downgrades_on_proof_wording_violation(self):
+        report = build_final_report(
+            {"proven": True},
+            {"complete": True},
+            {"claimable": True, "verdict": "strong"},
+            {"status": "clean"},
+            True,
+            False,
+            True,
+            report_text="Inventory complete after listing 4 open PRs.",
+        )
+        self.assertEqual(report["grade"], "downgraded")
+        self.assertFalse(report["proof_wording_proven"])
+        self.assertTrue(report["proof_wording_violations"])
+
+
+if __name__ == "__main__":
+    unittest.main()