READY-TO-ASSEMBLE-LEGO1-PATCH1 01 — Source Register, Codex HOLD Reconstruction & Mindset — 2026-06-22

Gate: REGISTRATION_HOLD · REGISTRATION_CAN_PROCEED = NO · 0 runtime mutations.

---

0. Mandatory mindset (recorded before any patch)

• I do not trust my READY report.
• I do not trust category completeness (Codex confirmed categories PASS but technical detail FAIL).
• I inspect actual governed files and actual planned commands.
• I reject any preparation package that uses prohibited raw DDL/DML.
• I reject overclaim that authority/execution are the only residuals if any design/preparation input remains.
• I must find the same bugs Codex found before reporting; Codex is for confirmation, not for discovering my bugs.

1. Direct-read register (no subagents)

Per Codex §3 caveat (the project rule forbids background agents/subagents, and file 01 of the prior package improperly cited subagent reads), PATCH1 used no subagents — every source below was read directly in the main process via mcp__agent-data__* and live read-only query_pg/pg_schema/list_docker.

Source	rev/len	proves
codex/codex-review-ready-to-assemble-lego1-preparation-package-2026-06-22.md	rev1, len 11399, truncated=false	controlling HOLD; findings P0/P1/P2; closure set
ready-to-assemble-lego1/ package (15 docs)	rev1–2	the defective wordings/commands being patched
ssot/directus/directus-operating-rules.md	v1.2 (S107)	DOT-only schema rule; MCP CRUD → 403; Registry Wiring Rule (table_registry)
laws-new/newlaws/dot-manage/dot-usage-handbook.md	canonical	dispatcher fn_process_agent_api_dispatch(p_dot_code,p_correlation_id,p_actor,p_mode,p_write_observation,p_idempotency_root); plan/apply pairs; "cannot itself execute a DOT"
laws-new/newlaws/dot-manage/README.md	canonical	"Directus/Postgres/schema is DOT-only. No manual SQL, no psql, no generic collection creation."
live dot_config	—	iu_create.gateway.direct_insert_policy=block_after_guard; process_dot_runtime.dry_run_only=true
live dot_agent_api_contract	—	paired DOT contract (producer DRY_RUN ↔ verifier VERIFY_ONLY); no_mutation_assertion; verifier_schema
live dot_iu_command_catalog	—	governed dot_iu_create_collection (mutating, reversible); staging gateway commands
live dot_iu_runtime_lease	—	atomic lease surface (lease_name,lease_token uuid,expires_at)
live v_dotkg_realrun_preflight	—	governed real-run gates → REALRUN_BLOCKED_MULTI_GATE
live apr_action_types / process_axis_action_vocabulary	—	authority-backed value source (provenance _dot_origin; canon-mutating ⇒ requires_president)

2. Codex finding reconstruction (reproduced independently)

Code	Codex finding (reconstructed)	I reproduced it by	Closed in
P0-1	File 06 plans raw CREATE TABLE/INSERT/audit-insert/UPDATE — violates DOT-only/Directus-API rule; WRITE_PLANNED_NOT_RUN label does not make the future plan valid	reading file 06 §3 templates against SSOT v1.2 + dot_config.iu_create.gateway.direct_insert_policy=block_after_guard	file 02
P0-2	operation_code text PRIMARY KEY cannot represent the same code across protocol_versions; successor_code not versioned; rollback updates by operation_code → wrong version	reading file 04 §1 schema + file 08 §5 rollback SQL	file 03
P1-1	Value set deferred to build = PREPARATION_INPUT_MISSING, not AUTHORITY_MISSING_ONLY; Gate B needs a reviewed hashed manifest or deterministic resolver	reading file 12 §2 / file 13 §2 claims vs absence of any value source	file 04
P1-2	Columns ≠ enforcement; PF5 conceptual not executable; consume at S8 after writes → double-execute/partial-failure-without-consume	reading file 06 step order (S8) + file 09 PF5 + file 11 §4	file 05
P1-3	No transaction boundaries / compensation / retry / partial-failure; rollback allows :successor=NULL, no successor-exists/version-binding/atomic-audit	reading file 06 §2 + file 08 §5	file 06
P1-4	PF7 "not tailed" yet claimed complete; PF8 proves denial on postgres DB not the directus/DOT write path; PF5 not exact	reading file 09 PF7/PF8/PF5	file 07
P1-5	Matrix is prose not executable; weak count>0/"by construction"; missing concurrency/version/partial-failure/forged-authority/serialization/audit-failure cases	reading file 07 §1–§6	file 08
P1-6	Hash `a	b	c
P2-1	"no incoming carrier edge" ≠ no build dependency; "blast radius=0" false (new schema/metadata/audit/auth/lookup)	reading file 05 §3 + file 06 §6 "blast radius=0"	file 10
P2-2	Remove "blast radius zero / structurally complete / inventory complete / only authority/execution remains" until proven	reading files 02/11/13	file 10

3. What PATCH1 does NOT reopen

Carrier selection (C1) remains valid (Codex §3 PASS_WITH_CAVEATS — "C1 is a plausible first carrier"); the accepted upstream Job A / I1–I10 / RBP oracle / Gate A-B split are not reopened. PATCH1 corrects only the C1 execution/identity/manifest/verifier/failure-model/preflight/test/serialization/blast-radius defects.

4. Boundary attestation

Direct reads only; no subagents; no runtime mutation; no DDL/DML; no Directus write; no DOT execution. REGISTRATION_HOLD retained; CAN_PROCEED = NO; 0 runtime mutations.