feat: block public claims on incomplete source checks

scripts/render_benchmark_reproducibility.py

@@ -313,6 +313,9 @@ def public_claim_blockers(
     world_class_ready: bool,
     world_class_open_gap_count: int,
     world_class_ledger_pending_count: int,
+    world_class_source_check_count: int,
+    world_class_source_pass_count: int,
+    world_class_source_blocked_count: int,
 ) -> list[str]:
     blockers = []
     if not local_reproducibility_ready:
@@ -328,6 +331,16 @@ def public_claim_blockers(
             f"world-class evidence is not accepted yet ({world_class_open_gap_count} open gaps, "
             f"{world_class_ledger_pending_count} ledger pending)"
         )
+    if (
+        world_class_source_check_count == 0
+        or world_class_source_pass_count != world_class_source_check_count
+        or world_class_source_blocked_count > 0
+    ):
+        blockers.append(
+            "world-class source checks are not all accepted "
+            f"({world_class_source_pass_count}/{world_class_source_check_count} pass, "
+            f"{world_class_source_blocked_count} blocked)"
+        )
     return blockers
 
 
@@ -379,6 +392,9 @@ def build_report(skill_dir: Path, generated_at: str) -> dict[str, Any]:
         world_class_ready,
         world_class_open_gap_count,
         world_class_ledger_pending_count,
+        world_class_source_check_count,
+        world_class_source_pass_count,
+        world_class_source_blocked_count,
     )
     public_claim_ready = not claim_blockers
     return {
@@ -422,7 +438,7 @@ def build_report(skill_dir: Path, generated_at: str) -> dict[str, Any]:
             "ready": public_claim_ready,
             "scope": "public benchmark or world-class readiness claim",
             "blockers": claim_blockers,
-            "policy": "Local reproducibility can pass before public claims; public claims require provider evidence, human adjudication, clean release lock, and accepted world-class evidence.",
+            "policy": "Local reproducibility can pass before public claims; public claims require provider evidence, human adjudication, clean release lock, accepted world-class evidence, and complete source checks.",
         },
         "release_lock": release_lock,
         "evidence_bundle": evidence_bundle,

tests/verify_benchmark_reproducibility.py

@@ -147,11 +147,13 @@ def main() -> None:
         == payload["summary"]["world_class_source_check_count"]
     ), payload
     assert payload["summary"]["public_claim_ready"] is False, payload
-    assert payload["summary"]["public_claim_blocker_count"] >= 3, payload
+    assert payload["summary"]["public_claim_blocker_count"] >= 4, payload
     assert payload["public_claim"]["ready"] is False, payload["public_claim"]
     assert any("provider-backed" in item for item in payload["public_claim"]["blockers"]), payload["public_claim"]
     assert any("human blind-review" in item for item in payload["public_claim"]["blockers"]), payload["public_claim"]
     assert any("world-class evidence" in item for item in payload["public_claim"]["blockers"]), payload["public_claim"]
+    assert any("world-class source checks" in item for item in payload["public_claim"]["blockers"]), payload["public_claim"]
+    assert "complete source checks" in payload["public_claim"]["policy"], payload["public_claim"]
     headings = {item["heading"]: item["exists"] for item in payload["methodology"]["sections"]}
     assert headings["## Benchmark Types"], headings
     assert headings["## Failure Disclosure"], headings
@@ -183,6 +185,7 @@ def main() -> None:
     assert "public claim ready: `false`" in markdown, markdown
     assert "## Public Claim Boundary" in markdown, markdown
     assert "provider-backed model holdout evidence is incomplete" in markdown, markdown
+    assert "world-class source checks are not all accepted" in markdown, markdown
     assert "## Release Lock" in markdown, markdown
     assert "## Evidence Bundle" in markdown, markdown
     assert "reports/benchmark_methodology.md" in markdown, markdown