{
  "record_version": "1.2.0",
  "date": "2026-07-15",
  "status": "proposed - pending leadership sign-off (A1-A5) and owner calls (D1-D6)",
  "register_page": "decisions.html",
  "evidence_page": "evidence.html",
  "kernel": {
    "path": "KERNEL.md",
    "version": "0.3.1",
    "standing_rule": "Permanent constraints shape the design; capability numbers only fill measured, re-measurable slots, re-derived every model generation. Dated evidence sets no threshold in either direction."
  },
  "recommendation": {
    "authorization_to_learn": "GO - one-project measurement pilot: baseline, contract compilation, adjudicated gold, deployment-tier bakeoff, shadow scoring, human-loop experiment",
    "authorization_to_act": "NO - no autonomous enforcement below calibrated level B; decisions of significant detriment are taken by a human everywhere"
  },
  "capability_currency": {
    "deployment_tier": "Claude Fable / Opus 4.8 class; GPT-5.5 / 5.6 class",
    "published_deployment_tier_measurements_on_this_task": 0,
    "grade_distribution": {
      "C": 16,
      "A": 79,
      "B": 40
    }
  },
  "valid_outcomes": [
    "stop",
    "narrow",
    "assisted_only",
    "selective_graduation"
  ],
  "graduation_rule": "For every critical error class, the one-sided 95 percent upper confidence bound must be below its approved ceiling on an untouched, adjudicated test set; ceilings, minimum effective sample, multiplicity treatment, and stop rules are locked before the test is opened.",
  "sequencing_note": "Recommendation only: a stripped E2 probe (seeded praise and planted defects, scored by deployment-tier models) runs first, in parallel with the baseline. Nothing operational is authorized by this note.",
  "legal_gate": "Two-sided workforce census (EU platform-work scope; US AEDT and selection-procedure exposure) with counsel on both before any consequence attaches.",
  "authorizations": [
    {
      "id": "a1",
      "title": "Approve the standard: instruction satisfaction grounded in evidence",
      "status": "needs leadership sign-off",
      "evidence_ids": [
        "AJ-02",
        "AQ-03",
        "AQ-04",
        "F26-08"
      ]
    },
    {
      "id": "a2",
      "title": "Choose the project and fund its adjudicated gold set",
      "status": "needs leadership sign-off",
      "evidence_ids": [
        "CT-03",
        "HS-01",
        "HS-03",
        "RR-02"
      ]
    },
    {
      "id": "a3",
      "title": "Measure the reviewer baseline first",
      "status": "needs leadership sign-off",
      "evidence_ids": [
        "CG-02",
        "CG-04"
      ]
    },
    {
      "id": "a4",
      "title": "Authority boundaries: measurement broadly, enforcement narrowly and earned",
      "status": "needs leadership sign-off",
      "evidence_ids": [
        "CG-07",
        "CG-11",
        "CM-03",
        "CM-10",
        "CT-02",
        "DA-02"
      ]
    },
    {
      "id": "a5",
      "title": "The audit channel is a fixture, not a phase",
      "status": "needs leadership sign-off",
      "evidence_ids": [
        "AQ-10",
        "CG-13",
        "CG-14",
        "CM-02",
        "CT-03"
      ]
    }
  ],
  "open_decisions": [
    {
      "id": "d1",
      "title": "Enforcement weight at launch",
      "recommended": "advisory first; consequences per lane after seeded error rates exist",
      "evidence_ids": [
        "CG-07",
        "CG-19",
        "CM-05",
        "DA-03"
      ]
    },
    {
      "id": "d2",
      "title": "Rubric authority",
      "recommended": "rubric as law with fast versioned amendment",
      "evidence_ids": [
        "CT-04",
        "RR-05"
      ]
    },
    {
      "id": "d3",
      "title": "AI-assistance policy",
      "recommended": "content validity by default; per-project override",
      "evidence_ids": [
        "AQ-01",
        "AQ-02",
        "AQ-08",
        "CG-10"
      ]
    },
    {
      "id": "d4",
      "title": "Judge sourcing",
      "recommended": "rent both tiers; revisit at measured volume",
      "evidence_ids": [
        "AJ-01",
        "GF-01",
        "GF-02",
        "IP-08"
      ]
    },
    {
      "id": "d5",
      "title": "Pilot locale and legal floors",
      "recommended": "two-sided census plus counsel before consequences",
      "evidence_ids": [
        "CG-07",
        "CG-08",
        "CG-09",
        "CG-12",
        "DA-01",
        "DA-02"
      ]
    },
    {
      "id": "d6",
      "title": "The throughput dial",
      "recommended": "thresholds move, measurement never; stamped batches, disclosed",
      "evidence_ids": [
        "CG-12",
        "F26-03",
        "F26-09",
        "HS-08"
      ]
    }
  ],
  "standing_defaults": [
    {
      "id": "c1",
      "title": "Reasonable-expert support, three-way verdicts",
      "falsifier": "measured false-flag rate vs project error prices"
    },
    {
      "id": "c2",
      "title": "Checkability routing; quantified collapse held provisional",
      "falsifier": "E3 internal replication"
    },
    {
      "id": "c3",
      "title": "Evidence-only to adjudicators; full explanations to workers",
      "falsifier": "E6 A/B"
    },
    {
      "id": "c4",
      "title": "Statistics staged: corrected estimates at v1, psychometrics at phase 2",
      "falsifier": "sequencing, not selection"
    },
    {
      "id": "c5",
      "title": "Single-item pay denial treated as significant detriment",
      "falsifier": "counsel opinion or transposition texts, due 2026-12"
    }
  ],
  "never_authorized": [
    "silent or retroactive changes to the bar workers are judged against",
    "calibration from unadjudicated override streams",
    "scoring any person on agreement with the machine or consensus",
    "stakes or leniency language in judge prompts",
    "raw percent-agreement in reporting",
    "verified labels on absence claims without seeded-recall evidence"
  ]
}