KB-2A56

Codex Recheck 3 Failure Matrix

4 min read Revision 1
fix7t1recheck-3failure-matrix2026-06-09

01 - Codex Recheck 3 Failure Matrix

Source: codex-fix7-blueprint-recheck-3-after-option-beta-patch-2026-06-08/ (00..09) + checkpoint, status FIX7_REFACTOR_BLUEPRINT_CODEX_RECHECK_3_NEEDS_T1_FIX.

Codex per-check verdicts → T1 disposition

Codex check (recheck-3 doc) Codex verdict T1 disposition this pass
A disposition removal (01) DISPOSITION_MODEL_REMOVAL_NEEDS_FIX (consistency only — guard scope vs history) FIXED via the ACTIVE_AUTHORITY boundary (blocker 2); doc 07
B U_legacy uniform end-state (02) U_LEGACY_UNIFORM_ENDSTATE_NEEDS_FIX (mixed-type denominator) FIXED via the three typed sets (blocker 1); docs 02/03/04/05
C STUB/body removal (03) STUB_BODY_PATH_REMOVED_ACCEPTED accepted; preserved (no change)
D owner isolation / #21 (04) OWNER_ISOLATION_21_CONTRACT_NEEDS_FIX (principal universe mixed into object denominator) FIXED — privilege is the object × principal tuple join (blocker 1); doc 03
E DO_NOT_TOUCH authority (05) DO_NOT_TOUCH_AUTHORITY_REMOVED_ACCEPTED accepted; non-PG DOT/scheduler moved to U_entry_vector (doc 04)
F hardcode (06) HARDCODE_NEEDS_FIX (contract precision: mixed set + ambiguous guard input) FIXED via set separation + boundary; doc 08
G PG-native (06) PG_NATIVE_DRIVEN_NEEDS_FIX (same root) FIXED; doc 08
H invariants/boundary (07) INVARIANTS_BOUNDARY_ACCEPTED preserved (27/11/14/7; the new boundary is doc metadata, not a DB surface); doc 09
I authoring planning (08) AUTHORING_PLANNING_NEEDS_T1_FIX unblocked by the two fixes; doc 07/09

The two blocker classes (verbatim intent)

Blocker 1 — set separation. "U_legacy is a PG object set, but the denominator currently mixes PG objects with principals and DOT/scheduler entry-vectors. This breaks the uniform end-state because principals/DOT/scheduler cannot satisfy owner/body/#21 object-state proof." Required: three distinct sets — object (U_legacy_object), principal (U_effective_privilege_principal), entry-vector (U_entry_vector); object both-EXCEPT only over PG objects; privilege checks join object × principal; bypass checks map entry-vector → object reachability, not object membership.

Blocker 2 — superseded-history boundary. "Guard no-disposition scans the entire blueprint, but historical sections still contain old disposition/STUB instructions. These sections are history, not active authority." Required: a machine-readable ACTIVE_AUTHORITY vs SUPERSEDED_NON_AUTHORITY boundary; the no-disposition guard must scan ACTIVE_AUTHORITY, report (not fail on) fenced history, fail on old terms in ACTIVE_AUTHORITY, fail if a future package references a superseded instruction, and fail if the boundary is absent/ambiguous/Directus-editable.

What Codex explicitly accepted (not reopened)

  • STUB / body-mutation path removed (legacy rollback = forward-only owner/ACL snapshot replay).
  • DO_NOT_TOUCH no longer subtracts from the legacy set (collisions fail closed).
  • 27 authority surfaces / 11 runtime-evidence tables / 14 readiness gates / 7 hash contracts intact.
  • Implementation, Stage 2.6B, permit, REAL_RUN, QT001 apply, activation, repoint, cutover BLOCKED.
  • Option Beta is the correct design and is not reopened or redesigned this pass.
Back to Knowledge Hub knowledge/dev/reports/architecture/t1-fix7-blueprint-patch-after-codex-recheck-3-set-separation-2026-06-08/01-codex-recheck-3-failure-matrix.md