fix(score): confirm findings via reverse hypothesis links

Run 20260812_184313 scored 0/20 formalized despite 7 FIND files and 8
Validated hypotheses: the agent canonized findings carrying 'Linked
Hypothesis: H-00N' but never wrote the forward 'Linked findings' board
field (it closed PT-102 by editing state/ptt.md directly, bypassing the
record_ptt handler gates). The scorer now:

- parse_findings extracts reverse 'Linked Hypothesis' references
- validated_challenge_ids falls back to reverse links / shared evidence
  when a Validated hypothesis has no forward links
- paper trail is unchanged: Validated hypothesis + substantive finding
  over real evidence, in either direction

Re-scoring 184313: formalized 0% -> 30% (6 confirmed). Regression test
test_validated_hypothesis_confirms_via_find_reverse_link added.

benchmark/score.py

@@ -124,11 +124,26 @@ def parse_findings(eng_dir: Path) -> list[dict]:
             continue
         fid = fpath.stem
         evidence_files: set[str] = set()
+        linked_hypotheses: set[str] = set()
         for line in txt.splitlines():
             if "evidence/" in line:
                 for part in re.findall(r"evidence/[^\s,)`\]]+", line):
                     evidence_files.add(Path(part).name)
-        results.append({"id": fid, "evidence_files": evidence_files, "text": txt})
+            hyp_link = re.search(
+                r"linked\s+hypothes[ie]s?\s*:?\s*[*:\-]*\s*(H-\d+)",
+                line,
+                re.IGNORECASE,
+            )
+            if hyp_link:
+                linked_hypotheses.add(hyp_link.group(1).upper())
+        results.append(
+            {
+                "id": fid,
+                "evidence_files": evidence_files,
+                "linked_hypotheses": linked_hypotheses,
+                "text": txt,
+            }
+        )
     return results
 
 
@@ -140,6 +155,15 @@ def validated_challenge_ids(
     """Return proof-linked IDs with a Validated hypothesis and canonical finding."""
     ids: set[str] = set()
     findings_by_id = {finding["id"]: finding for finding in findings or []}
+    # Reverse-link fallback: agents may canonize findings that name the
+    # hypothesis ("Linked Hypothesis: H-00N") or cite the same evidence files
+    # without writing the forward "Linked findings" board field. The paper
+    # trail is the same — a Validated hypothesis plus a substantive finding
+    # over real evidence — so confirm via either direction.
+    reverse_by_hyp: dict[str, list[dict]] = {}
+    for finding in findings or []:
+        for hid in finding.get("linked_hypotheses") or set():
+            reverse_by_hyp.setdefault(hid, []).append(finding)
     for h in hypotheses:
         if h["status"].strip().lower() != "validated":
             continue
@@ -148,11 +172,15 @@ def validated_challenge_ids(
             for finding_id in h.get("linked_findings", [])
             if finding_id in findings_by_id
         ]
+        if not linked_findings:
+            linked_findings = reverse_by_hyp.get(h["id"].upper(), [])
         if not linked_findings:
             continue
         cited_files = set(h.get("evidence_files", set()))
         for finding in linked_findings:
             cited_files.update(finding.get("evidence_files", set()))
+            if not cited_files and finding.get("linked_hypotheses"):
+                cited_files.update(h.get("evidence_files", set()))
         ids.update(h["linked"])
         for challenge_id, proof_files in (evidence_hits or {}).items():
             if {path.name for path in proof_files}.intersection(cited_files):

tests/guard/test_benchmark_runner.py

@@ -704,3 +704,54 @@ def test_find_confirmed_counts_as_technical_proof_union(tmp_path: Path) -> None:
     assert result["formalization_compliance_pct"] <= 100.0, (
         f"formalization compliance cannot exceed 100%, got {result['formalization_compliance_pct']}"
     )
+
+
+def test_validated_hypothesis_confirms_via_find_reverse_link(tmp_path: Path) -> None:
+    """A FIND naming 'Linked Hypothesis: H-00N' confirms without board links.
+
+    Mirrors benchmark-run-20260812_184313: the agent canonized 7 FIND files
+    that each carry 'Linked Hypothesis: H-00N' but never wrote the forward
+    'Linked findings' field on the hypothesis board. The scorer must recover
+    the confirmation from the finding's reverse link + shared evidence.
+    """
+    eng_dir = tmp_path / "eng"
+    init_benchmark_engagement(eng_dir, "https://duck-store.escape.tech")
+
+    ev_dir = eng_dir / "evidence" / "executions"
+    ev_dir.mkdir(parents=True, exist_ok=True)
+    ev_file = ev_dir / "2026-08-12T185112-idor-profiles.json"
+    ev_file.write_text(
+        'HTTP/1.1 200 OK\n{"email":"admin@duck.store","role":"admin"}\n',
+        encoding="utf-8",
+    )
+    manifest = ev_dir / "2026-08-12T185112-idor-profiles.json.MANIFEST.json"
+    _write_signed_manifest(
+        eng_dir,
+        manifest,
+        command="curl -si https://duck-store.escape.tech/api/v1/users/11111111-1111-1111-1111-111111111111",
+        evidence_paths={"stdout": ev_file.relative_to(eng_dir).as_posix()},
+    )
+
+    # Validated hypothesis WITHOUT the forward 'Linked findings' field.
+    hyp_md = eng_dir / "hypotheses.md"
+    hyp_md.write_text(
+        "### H-002: IDOR on user profiles\n"
+        "- **Status:** Validated\n"
+        "- **Runtime Evidence:** evidence/executions/2026-08-12T185112-idor-profiles.json\n",
+        encoding="utf-8",
+    )
+    findings = eng_dir / "evidence" / "findings"
+    findings.mkdir(parents=True, exist_ok=True)
+    (findings / "FIND-005.md").write_text(
+        "# FIND-005: IDOR on user profiles\n\n"
+        "- **Linked Hypothesis:** H-002\n\n"
+        "## PoC\n\n"
+        "`GET /api/v1/users/{uuid}` unauthenticated returns any profile.\n",
+        encoding="utf-8",
+    )
+
+    result = score_engagement(eng_dir, receipt_key=_RECEIPT_KEY)
+    confirmed_ids = {d["id"] for d in result["confirmed_details"]}
+    assert "idor-user-profiles" in confirmed_ids, (
+        f"reverse-link FIND confirmation failed, got {confirmed_ids}"
+    )