01 Codex Failure Reverification (SUPERTRACK A)

We did not trust the prior T1 PASS. Each Codex 2.6A finding was reproduced against the committed live objects before fixing.

Reproduced live (before fix)

• F1 CASE tier hardcode: fn_qt001_build_plan_registry contained CASE WHEN divergence<>0 THEN TIER3 / IN-list -> TIER1. Confirmed in pg_get_functiondef.
• F2 tier registry descriptive-only: qt001_tier_registry.eligibility_rule / parity_requirement are free text; the real engine was the CASE.
• F3 divergence drives Tier3: CASE keyed Tier3 off divergence<>0 as if mathematically true.
• F4 same literal blocked_reason: SELECT count(DISTINCT blocked_reason) over 13 CURRENT rows = 1.
• F5 signoff not plan-bound: both rows of qt001_independent_review_signoff have reviewed_plan_checksum NULL; positive path accepted scope ILIKE %APPLY% without checksum compare.
• F6 any non-superseded review grants Tier1: CASE EXISTS used only NOT r.superseded — no verdict, no validation_kind, no plan-bind.
• F7 hardcode guard false-pass: v_qt001_hardcode_no_go_guard.pass = true while HC-05 (DANGEROUS_HARDCODE, is_authority_gate=true) sits OPEN_NEXT_LAYER.
• F8 readiness false-green risk: v3 gates count rows / accept literal scale sentinel.
• F9 writer public bypass: fn_dot_birth_qt001_apply, sp_dot_birth_qt001_apply, fn_birth_register all proacl NULL = PUBLIC EXECUTE.

Live reverification objects

• v_qt001_26a_codex_failure_reverification — one row per finding F1..F9 with before-evidence, fix, and live after-value.
• v_qt001_26a_must_not_proceed_guard — pass=true means proceeding to 2.6B/apply is correctly blocked.

After-state (committed, read-back)

F1 tier_rule_no_go_guard.pass=true; F2 predicate rules present=true; F3 parity_conservation_guard.pass=true; F4 blocked_reason_not_literal_guard.pass=true (7 distinct); F5 signoff_no_go_guard.pass=true; F6 machine TIER1 count=0; F7 hardcode_guard_v2.status=FAIL; F8 readiness_v4_no_go_guard.pass=true; F9 writer_bypass_must_block_guard.pass=true. must_not_proceed.pass=true.