KB-2CCA

READY-TO-ASSEMBLE-LEGO1-PATCH2 01 — Source Register & Codex Failure-Reproduction Ledger — 2026-06-22

8 min read Revision 1
ready-to-assemble-lego1-patch2failure-reproduction-firstsource-registercodex-hold-reproductionno-subagentslive-readonly-evidenceregistration-hold2026-06-22

READY-TO-ASSEMBLE-LEGO1-PATCH2 01 — Source Register & Codex Failure-Reproduction Ledger — 2026-06-22

Gate: REGISTRATION_HOLD · CAN_PROCEED = NO · 0 runtime mutations. No subagents used (project rule); every read done in the main process. Live probes were all read-only SELECT/catalog/function-definition/view reads.


FAILURE_REPRODUCTION_FIRST

I do not trust my prior READY report. I do not trust category completeness. I do not trust harness names. I do not trust planned commands. I read actual governed files. I inspect actual live read-only surfaces. I reproduce Codex findings before patching. I create bad inputs beyond the provided matrix. If invalid input can still produce PASS/digest/seal/ready, I reject. I distinguish engineering PASS from authority PASS. I distinguish preparation-complete from dry-run authorization. I distinguish dry-run authorization from dry-run execution.

Order followed: (1) read the official Codex HOLD; (2) reproduce each finding from live read-only evidence; (3) prove each finding real or not; (4) only then design corrections; (5) run an internal Codex-style negative review (file 11); (6) report READY only if that review would PASS — it does not, so the verdict is HOLD.

1. Sources read directly

  • Controlling HOLD: knowledge/dev/laws-new/reports/codex/codex-review-ready-to-assemble-lego1-patch1-dry-run-auth-readiness-2026-06-22.md (rev 1, 21336 chars) — HOLD / NEED_READY_TO_ASSEMBLE_LEGO1_PATCH2.
  • PATCH1 package (all 14 docs), read in full: files 02–06, 08, 09 + internal self-review read verbatim; 01/07/10/11/index/codex-packet covered via the Codex review's per-file critique.
  • PATCH1 rollup: …/reports/macro-ready-to-assemble-lego1-patch1-2026-06-22.md (rev 3). Not trusted (held).
  • Accepted base / prior: …/codex/codex-review-ready-to-assemble-lego1-preparation-package-2026-06-22.md; …/codex/codex-review-rs5b-closeout-patch2-lego-readiness-seal-2026-06-21.md.

2. Live read-only discovery (VPS directus, PostgreSQL 16, this session)

Every probe below was executed by me this session via query_pg (read-only role, READ ONLY txn, LIMIT 500) — independent of Codex.

Probe Command (summary) Result READ_ONLY_SAFE
Dispatcher def pg_get_functiondef(fn_process_agent_api_dispatch) contains IF p_mode='REAL_RUN' THEN RAISE EXCEPTION 'dispatch refused: REAL_RUN not permitted by this dispatcher'; note "dispatcher cannot execute a DOT and never writes DRY_RUN/REAL_RUN" yes
Create-collection def pg_get_functiondef(fn_iu_collection_create) INSERT INTO public.iu_piece_collection (...); gated by fn_iu_composer_enabled() yes
Contract catalog SELECT … FROM dot_agent_api_contract 2 rows only: DOT_KG_EXPLAIN(+_VERIFY); endpoint http://incomex-agent-api-executor:8090/dispatch yes
IU command catalog SELECT … FROM dot_iu_command_catalog 54 commands, all over iu_piece_collection; none create Directus collections/fields/physical tables/table_registry yes
Real-run preflight SELECT * FROM v_dotkg_realrun_preflight DOT_KG-specific gates; OVERALL_VERDICT=REALRUN_BLOCKED_MULTI_GATE / NO_GO yes
Auth constraints pg_get_constraintdef on governance_build_authorization status IN (draft,active,consumed,expired,revoked); chk_sov_esign = presence-only yes
Auth/resolver counts one combined SELECT apr_active=14, pav_total=12, join(action_code)=0, auth_count=0 yes
Harness def pg_get_functiondef(fn_iu_bcf_harness_run) tests IU axis B/C/F with hardcoded params; calls fn_dot_iu_command_log ×5 yes
Config SELECT … FROM dot_config dry_run_only=true, execute_enabled=false, real_run_enabled=false, iu_core.composer_enabled=false, iu_create.gateway.direct_insert_policy=block_after_guard yes
C1 surface combined existence SELECT c1_table=0, c1_dot_tools=0, c1_contracts=0, c1_preflight_views=0 yes

3. Codex failure-reproduction ledger (R1–R12)

Legend — Falsely-passed? = would the PATCH1 self-review have reported PASS for this defect.

ID Codex finding PATCH1 source Live evidence (mine) Reproduced REAL? Falsely-passed? Closure target
R1 Dispatcher does not execute REAL_RUN file 02 §4 (G4/G5), file 06 §2 model REAL_RUN via dispatcher dispatcher RAISES on REAL_RUN; "never writes DRY_RUN/REAL_RUN" YES YES file 02: REAL_RUN via executor endpoint + governed promotion; dispatcher = PLAN/VERIFY/DRY_RUN only
R2 dot_iu_create_collection ≠ Directus schema file 02 §2/§4, file 06 §2 fn_iu_collection_createiu_piece_collection; composer gate closed; no Directus-DDL primitive in 54-cmd catalog YES YES file 02: new governed DOT_C1_SCHEMA_ENSURE (Form B)
R3 C1 DOT pair absent file 02 §3 names DOT_C1_VOCAB_BUILD c1_contracts=0, c1_dot_tools=0, c1_table=0 YES YES file 02: register producer/verifier pair (absent → prep gap)
R4 PF7 uses wrong (DOT_KG) contract file 07 §3 calls v_dotkg_realrun_preflight "the C1 build's governing preflight" view gates are DOT_KG-specific; c1_preflight_views=0 YES YES file 07: C1-specific v_c1_realrun_preflight (Form B)
R5 Verifier requires impossible status granted file 05 §2 g.status='granted' status domain excludes granted; auth_count=0 YES YES file 05: status='active' (real lifecycle active→consumed)
R6 Verifier matching not exact file 05 §2 action_set @> :req @> permits supersets YES (design) YES file 05: exact set-equality + sorted canonical compare
R7 Action superset risk file 05 §2 same @> defect YES (design) YES file 05: equality, not containment
R8 Resolver join 0 / provenance≠authority file 04 §2 enrich via process_axis_action_vocabulary join(action_code)=0; the two tables are different namespaces (create_item/add_field vs APPROVE_BIRTH_ADMISSION/CONFIRM_AS_PROCESS) YES YES file 04: drop broken join; authority from apr_action_types' own columns
R9 Invoked harness unrelated to C1 file 08 §2 "dot_iu_test_harness_run may run the suite" fn_iu_bcf_harness_run tests IU axis B/C/F only YES YES file 08: C1 harness contract (Form B; absent → prep gap)
R10 cser lacks canonical bytes + recompute file 09 (no bytes/digest/command) n/a (doc-level) YES YES file 09: real bytes + SHA-256 + recompute command (CLOSED)
R11 Compensation labels not executable file 06 §2 COMP_* are labels reversible lifecycle DOTs exist but only for IU pieces, not C1 YES YES file 06: governed C1 compensation handlers (absent → prep gap)
R12 Blast radius omits to-create artifacts file 10 §1–4 the schema DOT/contract/harness/preflight/consume handler all absent YES YES file 10: truthful map including all absent artifacts

Result: every Codex finding reproduced as REAL from independent live evidence. None was a Codex error. The PATCH1 self-review (file 11 of PATCH1) marked all of them PASS — it was a category-completeness checklist, not an adversarial live-evidence review; this is the root of B12 and is corrected in file 11 here.

4. Boundary attestation

Reproduction only; no mutation, no DOT executed, no write. REGISTRATION_HOLD retained; CAN_PROCEED = NO; 0 runtime mutations.

Back to Knowledge Hub knowledge/dev/laws-new/reports/ready-to-assemble-lego1-patch2/01-source-register-and-codex-failure-reproduction-2026-06-22.md