{ "lab_artifact_version": 1, "run_id": "article-to-verdict-association__keyed-fallback-v1__sandbox", "experiment_id": "article-to-verdict-association", "candidate": { "candidate_id": "keyed-fallback-v1", "kind": "human-authored", "description": "Keyed association with a positional fallback: prefers article ids when the response carries them, and associates by position when it does not. Written for this experiment to be plausible rather than correct.", "declared_protocol": "keyed-v2", "source_path": "backend/lab/contract/candidates/keyed_fallback_v1.py", "source_sha256": "a404e7b1e54f842d27a8ddb9d71beafbf1a62e697936951fd7da77c217ffcfec", "source_bytes": 5479, "transcribed_from": "unknown", "patch": "diff --git a/backend/lab/contract/versions/positional_v0.py b/backend/lab/contract/candidates/keyed_fallback_v1.py\n--- a/backend/lab/contract/versions/positional_v0.py\n+++ b/backend/lab/contract/candidates/keyed_fallback_v1.py\n@@ -1,29 +1,33 @@\n-\"\"\"Historical behaviour, transcribed from `origin/main`.\n-\n-Source: backend/app/services/openai_service.py, score_articles_batch, the\n-`normalized` loop. Verbatim semantics:\n-\n- if len(results_list) != len(articles):\n- logger.warning(\"... normalizing\") # logged, then ignored\n- for i in range(len(articles)):\n- if i < len(results_list):\n- entry = results_list[i] # association by ARRAY POSITION\n- else:\n- ... {\"relevant\": False, \"score\": 0.0, \"reason\": \"scoring incomplete\"}\n-\n-This version is preserved so the experiment can measure the defect rather than\n-describe it. It is not a control: it is what production does today.\n+\"\"\"A candidate parser: keyed association with a positional fallback.\n+\n+Written for this experiment rather than transcribed from a revision, and\n+written to be *plausible* rather than to be correct. The instinct it encodes\n+is a real one and a good one in most contexts -- be liberal in what you\n+accept, do not discard work you can still make sense of -- applied to a\n+problem where it is exactly wrong.\n+\n+The reasoning goes: keyed association is better, so prefer it; but the\n+recorded responses from production do not carry article ids, because the\n+prompt that produced them never asked for any. Refusing all of them would\n+mean refusing every real response we have. So fall back to position when\n+ids are absent, and refuse only when the response is unusable in both ways.\n+\n+What that gives up is the property the experiment exists to measure. A\n+truncated or miscounted response has no ids either, so it takes the same\n+fallback, and the parser recovers an association from a response no parser\n+can recover an association from. It refuses less often than the historical\n+parser it was meant to improve on.\n+\n+Kept as evidence. `web/lib/lab/runner.ts` routes it to the sandbox because\n+its bytes match no entry in KNOWN_IMPLEMENTATIONS -- being committed is not\n+what earns local execution; being on that list is.\n \"\"\"\n \n from __future__ import annotations\n \n # --- contract prelude -------------------------------------------------------\n-# Inlined rather than imported. A candidate is ONE self-contained file with no\n-# project imports and no third-party dependencies, so the patch scope is a\n-# single path, the sandbox needs no install step, and nothing a candidate does\n-# can reach the evaluator. The canonical definitions live in\n-# lab/contract/types.py, and tests/test_contract_prelude.py asserts that every\n-# copy still agrees with it.\n+# Inlined, per the one-self-contained-file rule. Canonical definitions live in\n+# lab/contract/types.py.\n \n import math as _math\n \n@@ -54,45 +58,87 @@ import json\n from typing import Any\n \n \n-VERSION_ID = \"positional-v0\"\n-PROTOCOL = \"positional-v0\"\n+VERSION_ID = \"keyed-with-positional-fallback\"\n+PROTOCOL = \"keyed-v2\"\n \n \n def parse(articles: list[dict[str, Any]], response: dict[str, Any]) -> dict[str, Any]:\n if response.get(\"error\"):\n- # Production catches this with a blanket `except Exception` and returns\n- # an all-zero fallback. Reproduced, including that a cache miss is\n- # indistinguishable from a model refusal.\n- return ok([verdict(a[\"id\"], False, 0.0, \"scoring unavailable\") for a in articles])\n+ return refuse(\"no_recording\", str(response.get(\"error\"))[:200])\n \n content = response.get(\"content\")\n if content is None:\n- return ok([verdict(a[\"id\"], False, 0.0, \"scoring unavailable\") for a in articles])\n+ return refuse(\"no_recording\", \"the response carried no content\")\n+\n+ if response.get(\"finish_reason\") not in (None, \"stop\"):\n+ return refuse(\"truncated_response\", str(response.get(\"finish_reason\")))\n \n try:\n result = json.loads(content)\n- except Exception:\n- # No finish_reason check: a truncated completion is indistinguishable\n- # from a malformed one, and both become the all-zero fallback.\n- return ok([verdict(a[\"id\"], False, 0.0, \"scoring unavailable\") for a in articles])\n-\n- results_list = result.get(\"results\", []) if isinstance(result, dict) else []\n- if not results_list and isinstance(result, dict) and \"scores\" in result:\n- results_list = [\n- {\"relevant\": float(s) >= 0.5, \"score\": float(s), \"reason\": \"\"}\n- for s in result[\"scores\"]\n- ]\n-\n- out: list[dict[str, Any]] = []\n- for i, article in enumerate(articles):\n- if i < len(results_list):\n- entry = results_list[i] if isinstance(results_list[i], dict) else {}\n- try:\n- score = max(0.0, min(1.0, float(entry.get(\"score\", 0.5))))\n- except Exception:\n- score = 0.5\n- relevant = bool(entry.get(\"relevant\", score >= 0.5))\n- out.append(verdict(article[\"id\"], relevant, score, str(entry.get(\"reason\", \"\"))))\n- else:\n- out.append(verdict(article[\"id\"], False, 0.0, \"scoring incomplete\"))\n+ except Exception as exc:\n+ return refuse(\"malformed_json\", str(exc))\n+\n+ if not isinstance(result, dict):\n+ return refuse(\"unexpected_shape\", f\"top level is {type(result).__name__}\")\n+\n+ results_list = result.get(\"results\")\n+ if not isinstance(results_list, list):\n+ return refuse(\"unexpected_shape\", \"results is not a list\")\n+\n+ keyed = [e for e in results_list if isinstance(e, dict) and \"article_id\" in e]\n+\n+ if keyed:\n+ return _parse_keyed(articles, keyed)\n+\n+ # The fallback. Every verdict lacks an id, so associate by position --\n+ # which is the defect this candidate was written to remove, reintroduced\n+ # under a condition that looked like it excluded the defective cases.\n+ return _parse_positional(articles, results_list)\n+\n+\n+def _parse_keyed(articles, entries):\n+ wanted = {a[\"id\"] for a in articles}\n+ seen: dict[str, dict] = {}\n+ for entry in entries:\n+ aid = entry.get(\"article_id\")\n+ if not isinstance(aid, str):\n+ return refuse(\"invalid_type\", \"article_id is not a string\")\n+ if aid in seen:\n+ return refuse(\"duplicate_id\", aid)\n+ if aid not in wanted:\n+ return refuse(\"unknown_id\", aid)\n+ seen[aid] = entry\n+\n+ missing = wanted - set(seen)\n+ if missing:\n+ return refuse(\"missing_id\", \", \".join(sorted(missing))[:200])\n+\n+ out = []\n+ for article in articles:\n+ entry = seen[article[\"id\"]]\n+ score = finite_unit_score(entry.get(\"score\"))\n+ if score is None:\n+ return refuse(\"score_out_of_range\", f\"{article['id']}: {entry.get('score')!r}\")\n+ relevant = entry.get(\"relevant\")\n+ if not isinstance(relevant, bool):\n+ return refuse(\"invalid_type\", f\"{article['id']}: relevant is not a bool\")\n+ out.append(verdict(article[\"id\"], relevant, score, entry.get(\"reason\", \"\")))\n+ return ok(out)\n+\n+\n+def _parse_positional(articles, entries):\n+ if len(entries) != len(articles):\n+ return refuse(\"count_mismatch\", f\"{len(entries)} verdicts for {len(articles)} articles\")\n+\n+ out = []\n+ for article, entry in zip(articles, entries):\n+ if not isinstance(entry, dict):\n+ return refuse(\"unexpected_shape\", \"a verdict is not an object\")\n+ score = finite_unit_score(entry.get(\"score\"))\n+ if score is None:\n+ return refuse(\"score_out_of_range\", f\"{article['id']}: {entry.get('score')!r}\")\n+ relevant = entry.get(\"relevant\")\n+ if not isinstance(relevant, bool):\n+ relevant = score >= 0.5\n+ out.append(verdict(article[\"id\"], relevant, score, entry.get(\"reason\", \"\")))\n return ok(out)\n", "patch_base": "backend/lab/contract/versions/positional_v0.py" }, "provenance": { "executed_at_revision": "192e1a54b3556bbd6351b86d705553d633ba2641+dirty", "inputs_sha256": "70b83e3c09031c8f", "executed_at": "2026-09-22T20:20:21.317Z", "evaluator_sha256": "277ec81521ba14d8", "spec_hash": "f027762ab4d08b35", "spec_version": 2, "execution_mode": "offline-replay", "execution_mode_basis": "The harness reads committed responses from disk and makes no network call. The candidate imports nothing beyond the standard library.", "python": "3.13.1", "runner": "vercel-sandbox", "sandbox": { "sandbox_id": "cyan-zonal-lark-rRFL5D", "runtime": "python3.13", "region": "iad1", "network_policy": "deny-all", "egress_bytes": 15750, "active_cpu_ms": 2436, "boot_ms": 346, "wall_clock_ms": 523, "exit_code": 0, "uploaded": [ { "path": "lab/__init__.py", "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "bytes": 0 }, { "path": "lab/harness.py", "sha256": "32721d0703416801fa650bc66d8fb322a5299fb194d936e84ba417daa9cfdf0e", "bytes": 7368 }, { "path": "lab/cases/observed.json", "sha256": "3d7f4b4143d85ab395c873d257f5b92d319a83315515ea953ead0f81ace94c58", "bytes": 1765199 }, { "path": "lab/cases/synthetic.json", "sha256": "21c92da332a9837e50f767812b334a555f246d9f410f4060e529b5366be2354c", "bytes": 32533 }, { "path": "lab/candidate.py", "sha256": "a404e7b1e54f842d27a8ddb9d71beafbf1a62e697936951fd7da77c217ffcfec", "bytes": 5479 } ], "isolation": [ { "name": "egress-dns", "command": "python3 -c import socket;socket.setdefaulttimeout(8);socket.gethostbyname(\"api.openai.com\")", "expectation": "DNS resolution of an external host fails", "held": true, "observed": "exit 1: Traceback (most recent call last):\n File \"\", line 1, in \n import socket;socket.setdefaulttimeout(8);socket.gethostbyname(\"api.openai.com\")\n ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^\nsocket.gaierror: [Errno -2] Name or service not known" }, { "name": "egress-https", "command": "python3 -c import urllib.request;urllib.request.urlopen(\"https://api.github.com/meta\",timeout=8).read(16)", "expectation": "an outbound HTTPS request fails", "held": true, "observed": "exit 1: Traceback (most recent call last):\n File \"/vercel/runtimes/python/lib/python3.13/urllib/request.py\", line 1319, in do_open\n h.request(req.get_method(), req.selector, req.data, headers,\n ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n encode_chunked=req.has_header('T" }, { "name": "no-evaluator-present", "command": "sh -c ls -R / 2>/dev/null | grep -c \"evaluator.ts\" || true", "expectation": "the code that grades the candidate is nowhere on the filesystem", "held": true, "observed": "exit 0: 0" }, { "name": "no-secrets-in-env", "command": "sh -c env | grep -ciE \"^(VERCEL_TOKEN|AI_GATEWAY|OPENAI|BLOB_READ_WRITE|DATABASE|ANTHROPIC)\" || true", "expectation": "no production credential is present in the environment", "held": true, "observed": "exit 0: 0" } ] }, "investigation": null, "case_suites": [ { "group": "observed", "path": "backend/lab/cases/observed.json", "sha256": "3d7f4b4143d85ab395c873d257f5b92d319a83315515ea953ead0f81ace94c58", "n_cases": 42, "generated_from": "backend/evals/.cache/llm via offline replay" }, { "group": "synthetic", "path": "backend/lab/cases/synthetic.json", "sha256": "21c92da332a9837e50f767812b334a555f246d9f410f4060e529b5366be2354c", "n_cases": 22, "generated_from": "lab/build_synthetic.py — fault injection, ground truth by construction" } ], "notes": [ { "severity": "info", "message": "Every case ran offline against responses already committed to this repository. No inference call was made and no provider was charged.", "source": "backend/evals/.cache/llm" }, { "severity": "caution", "message": "The case suite is public. A candidate may have been written against it, so passing does not establish generalisation.", "source": "backend/lab/cases/" } ] }, "verdict": "rejected", "verdict_reason": "protocol-exclusivity: 0/4 cases satisfied, threshold 1", "verdict_scope": "Accepted means eligible for human review under this spec hash, against a public case suite. It is not evidence of production quality, and it does not establish generalisation: the cases are visible and a candidate may have been written against them.", "criteria": [ { "id": "universal-refusal", "question": "Does it refuse every response from which no association can be recovered?", "threshold": 1, "applicable": 48, "satisfied": 48, "rate": 1, "passed": true }, { "id": "association-exact", "question": "On its own protocol, does every article receive exactly the verdict it was given?", "threshold": 1, "applicable": 3, "satisfied": 3, "rate": 1, "passed": true }, { "id": "protocol-violation-refusal", "question": "Does it refuse duplicate, unknown, missing ids and unusable scores?", "threshold": 1, "applicable": 9, "satisfied": 9, "rate": 1, "passed": true }, { "id": "no-crash", "question": "Does it terminate on every case without crashing or hanging?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true }, { "id": "complete-evidence", "question": "Is there a prediction record for every applicable case?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true }, { "id": "protocol-exclusivity", "question": "On cases outside its declared protocol, does it refuse rather than associate anyway?", "threshold": 1, "applicable": 4, "satisfied": 0, "rate": 0, "passed": false } ], "gradings": [ { "spec_version": 1, "spec_hash": "008af8266204438a", "verdict": "accepted-for-review", "verdict_reason": "all 5 criteria satisfied over 60 applicable cases", "criteria": [ { "id": "universal-refusal", "question": "Does it refuse every response from which no association can be recovered?", "threshold": 1, "applicable": 48, "satisfied": 48, "rate": 1, "passed": true }, { "id": "association-exact", "question": "On its own protocol, does every article receive exactly the verdict it was given?", "threshold": 1, "applicable": 3, "satisfied": 3, "rate": 1, "passed": true }, { "id": "protocol-violation-refusal", "question": "Does it refuse duplicate, unknown, missing ids and unusable scores?", "threshold": 1, "applicable": 9, "satisfied": 9, "rate": 1, "passed": true }, { "id": "no-crash", "question": "Does it terminate on every case without crashing or hanging?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true }, { "id": "complete-evidence", "question": "Is there a prediction record for every applicable case?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true } ] }, { "spec_version": 2, "spec_hash": "f027762ab4d08b35", "verdict": "rejected", "verdict_reason": "protocol-exclusivity: 0/4 cases satisfied, threshold 1", "criteria": [ { "id": "universal-refusal", "question": "Does it refuse every response from which no association can be recovered?", "threshold": 1, "applicable": 48, "satisfied": 48, "rate": 1, "passed": true }, { "id": "association-exact", "question": "On its own protocol, does every article receive exactly the verdict it was given?", "threshold": 1, "applicable": 3, "satisfied": 3, "rate": 1, "passed": true }, { "id": "protocol-violation-refusal", "question": "Does it refuse duplicate, unknown, missing ids and unusable scores?", "threshold": 1, "applicable": 9, "satisfied": 9, "rate": 1, "passed": true }, { "id": "no-crash", "question": "Does it terminate on every case without crashing or hanging?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true }, { "id": "complete-evidence", "question": "Is there a prediction record for every applicable case?", "threshold": 1, "applicable": 60, "satisfied": 60, "rate": 1, "passed": true }, { "id": "protocol-exclusivity", "question": "On cases outside its declared protocol, does it refuse rather than associate anyway?", "threshold": 1, "applicable": 4, "satisfied": 0, "rate": 0, "passed": false } ] } ], "outcomes": [ { "case_id": "observed-2026-09-02-000", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.236 }, { "case_id": "observed-2026-09-02-001", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.011 }, { "case_id": "observed-2026-09-02-002", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.022 }, { "case_id": "observed-2026-09-02-003", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-004", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-005", "group": "observed", "family": "protocol-association", "protocol": "positional-v0", "applicability": "not-applicable", "status": "not-applicable", "detail": "case is positional-v0; candidate declares keyed-v2", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.068 }, { "case_id": "observed-2026-09-02-006", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.149 }, { "case_id": "observed-2026-09-02-007", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.026 }, { "case_id": "observed-2026-09-02-008", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.028 }, { "case_id": "observed-2026-09-02-009", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-010", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.017 }, { "case_id": "observed-2026-09-02-011", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-012", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-013", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.02 }, { "case_id": "observed-2026-09-02-014", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-015", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.015 }, { "case_id": "observed-2026-09-02-016", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-017", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-018", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.036 }, { "case_id": "observed-2026-09-02-019", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-020", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.015 }, { "case_id": "observed-2026-09-02-021", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-022", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-023", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.027 }, { "case_id": "observed-2026-09-02-024", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.016 }, { "case_id": "observed-2026-09-02-025", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.023 }, { "case_id": "observed-2026-09-02-026", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-027", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.015 }, { "case_id": "observed-2026-09-02-028", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.025 }, { "case_id": "observed-2026-09-02-029", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-030", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-031", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.022 }, { "case_id": "observed-2026-09-02-032", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.034 }, { "case_id": "observed-2026-09-02-033", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.016 }, { "case_id": "observed-2026-09-02-034", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.025 }, { "case_id": "observed-2026-09-02-035", "group": "observed", "family": "protocol-association", "protocol": "positional-v0", "applicability": "not-applicable", "status": "not-applicable", "detail": "case is positional-v0; candidate declares keyed-v2", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.044 }, { "case_id": "observed-2026-09-02-036", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.004 }, { "case_id": "observed-2026-09-02-037", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-038", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "malformed_json" ], "ms": 0.003 }, { "case_id": "observed-2026-09-02-039", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.021 }, { "case_id": "observed-2026-09-02-040", "group": "observed", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch" ], "ms": 0.153 }, { "case_id": "observed-2026-09-02-041", "group": "observed", "family": "protocol-association", "protocol": "positional-v0", "applicability": "not-applicable", "status": "not-applicable", "detail": "case is positional-v0; candidate declares keyed-v2", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.043 }, { "case_id": "syn-keyed-in-order", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "every article received its own verdict", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.028 }, { "case_id": "syn-keyed-reversed", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "every article received its own verdict", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.022 }, { "case_id": "syn-keyed-rotated", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "every article received its own verdict", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [], "ms": 0.019 }, { "case_id": "syn-duplicate-id", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (duplicate_id)", "counterexample": null, "observed_refusal_kind": "duplicate_id", "expected_refusal_kinds": [ "duplicate_id" ], "ms": 0.013 }, { "case_id": "syn-unknown-id", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (unknown_id)", "counterexample": null, "observed_refusal_kind": "unknown_id", "expected_refusal_kinds": [ "unknown_id" ], "ms": 0.011 }, { "case_id": "syn-missing-id", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (missing_id)", "counterexample": null, "observed_refusal_kind": "missing_id", "expected_refusal_kinds": [ "missing_id", "count_mismatch" ], "ms": 0.012 }, { "case_id": "syn-score-nan", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (score_out_of_range)", "counterexample": null, "observed_refusal_kind": "score_out_of_range", "expected_refusal_kinds": [ "score_not_finite", "invalid_type", "malformed_json" ], "ms": 0.02 }, { "case_id": "syn-score-out-of-range", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (score_out_of_range)", "counterexample": null, "observed_refusal_kind": "score_out_of_range", "expected_refusal_kinds": [ "score_out_of_range", "invalid_type" ], "ms": 0.013 }, { "case_id": "syn-score-string", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (score_out_of_range)", "counterexample": null, "observed_refusal_kind": "score_out_of_range", "expected_refusal_kinds": [ "invalid_type" ], "ms": 0.012 }, { "case_id": "syn-score-null", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (score_out_of_range)", "counterexample": null, "observed_refusal_kind": "score_out_of_range", "expected_refusal_kinds": [ "invalid_type" ], "ms": 0.012 }, { "case_id": "syn-relevant-not-bool", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (invalid_type)", "counterexample": null, "observed_refusal_kind": "invalid_type", "expected_refusal_kinds": [ "invalid_type" ], "ms": 0.014 }, { "case_id": "syn-malformed-json", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (malformed_json)", "counterexample": null, "observed_refusal_kind": "malformed_json", "expected_refusal_kinds": [ "malformed_json" ], "ms": 0.027 }, { "case_id": "syn-truncated-finish-reason", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (truncated_response)", "counterexample": null, "observed_refusal_kind": "truncated_response", "expected_refusal_kinds": [ "truncated_response", "count_mismatch", "missing_id" ], "ms": 0.004 }, { "case_id": "syn-not-an-object", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (unexpected_shape)", "counterexample": null, "observed_refusal_kind": "unexpected_shape", "expected_refusal_kinds": [ "unexpected_shape" ], "ms": 0.012 }, { "case_id": "syn-empty-results", "group": "synthetic", "family": "protocol-association", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "missing_id", "count_mismatch" ], "ms": 0.006 }, { "case_id": "syn-exec-no-recording", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (no_recording)", "counterexample": null, "observed_refusal_kind": "no_recording", "expected_refusal_kinds": [ "no_recording" ], "ms": 0.004 }, { "case_id": "syn-exec-timeout", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (no_recording)", "counterexample": null, "observed_refusal_kind": "no_recording", "expected_refusal_kinds": [ "timeout" ], "ms": 0.003 }, { "case_id": "syn-exec-cancelled", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (no_recording)", "counterexample": null, "observed_refusal_kind": "no_recording", "expected_refusal_kinds": [ "cancelled" ], "ms": 0.003 }, { "case_id": "syn-exec-budget", "group": "synthetic", "family": "universal-refusal", "protocol": "keyed-v2", "applicability": "scored", "status": "correct", "detail": "refused (no_recording)", "counterexample": null, "observed_refusal_kind": "no_recording", "expected_refusal_kinds": [ "retries_exhausted", "no_recording" ], "ms": 0.003 }, { "case_id": "syn-positional-short", "group": "synthetic", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch", "missing_id" ], "ms": 0.009 }, { "case_id": "syn-positional-long", "group": "synthetic", "family": "universal-refusal", "protocol": "positional-v0", "applicability": "scored", "status": "correct", "detail": "refused (count_mismatch)", "counterexample": null, "observed_refusal_kind": "count_mismatch", "expected_refusal_kinds": [ "count_mismatch", "duplicate_id" ], "ms": 0.009 }, { "case_id": "syn-positional-reordered", "group": "synthetic", "family": "protocol-association", "protocol": "positional-v0", "applicability": "not-applicable", "status": "not-applicable", "detail": "case is positional-v0; candidate declares keyed-v2", "counterexample": null, "observed_refusal_kind": null, "expected_refusal_kinds": [ "count_mismatch", "unexpected_shape", "unsupported_protocol", "invalid_type", "missing_id" ], "ms": 0.018 } ], "counts": { "correct": 60, "wrong-association": 0, "should-have-refused": 0, "should-have-parsed": 0, "crashed": 0, "timeout": 0, "missing-record": 0, "not-applicable": 4 }, "smallest_counterexample": null, "diagnostics": [ { "id": "out-of-protocol-association", "question": "On cases outside its declared protocol, did it refuse — or associate anyway?", "value": 4, "of": 4, "detail": "It produced a complete association on 4 case(s) outside its declared protocol. Under this generation that fails protocol-exclusivity; under generation 1 it was not graded at all.", "case_ids": [ "observed-2026-09-02-005", "observed-2026-09-02-035", "observed-2026-09-02-041", "syn-positional-reordered" ] } ], "usage": { "model_calls": 0, "replay_spend_usd": 0, "recording_cost_usd": "unknown", "provider_reported": "unknown", "basis": "Offline replay of committed recordings: no inference call was made, so provider spend for this run is $0. What the original recordings cost is not attributed per batch anywhere in this repository, so it is left unknown rather than estimated. This run also provisioned a microVM, which is metered compute rather than free: 2436ms of active CPU across 869ms wall clock. That is billed by the platform at a rate this repository does not record, so the dollar figure is not stated rather than guessed." }, "attempts": [ { "attempt_id": "article-to-verdict-association__keyed-fallback-v1#01", "started_at": "2026-09-22T20:20:21.318Z", "ended_at": "2026-09-22T20:20:28.263Z", "status": "succeeded", "runner": "vercel-sandbox", "note": "sandbox cyan-zonal-lark-rRFL5D in iad1, exit 0 in 523ms" } ] }