KB-17F4

GPT Review — 18b DOT-119 Repair Execution Pack rev3

3 min read Revision 1
gpt-reviewdot-11918brev3p0-approvedumbrella-designdieu20

GPT Review — 18b DOT-119 Repair Execution Pack rev3

Date: 2026-05-05 Reviewer: GPT-5.5 Thinking / Incomex Hội đồng AI Reviewed: Opus draft 18b — DOT-119 Repair Execution Pack rev3 from conversation

Verdict

PASS as umbrella/gated execution design. Approve next step: draft P0 read-only runtime prompt only.

Rev3 fixes the rev2 blockers:

  • 18b is now explicitly an umbrella design, not a one-shot dispatch prompt.
  • P0 has a KB report path.
  • Schema/migration tool preflight is included.
  • HC registration is deferred by default.
  • DOT-119 script rewrite requires exact script content in P2, not Claude Code improvisation.
  • Exit codes align with 18a rev4.
  • IU is dry-run only in 18b; verify belongs to 18c.
  • Rollback dependency check is included.
  • Manual metadata provenance is included.

Remaining note

Do not dispatch P1/P2/P3 yet. The only approved next runtime action is P0 read-only mapping/preflight.

P0 must not mutate anything. It should only read and upload:

knowledge/dev/laws/dieu44-trien-khai/reports/dot119-repair-p0-readonly-mapping-report.md

After P0 report, GPT/User must review before any DDL/seed/script work.

P0 prompt requirements

Opus should now draft a compact Claude Code prompt for P0 only:

  1. Read controlling docs:

    • 18a-dot119-metadata-driven-no-clobber-redesign.md rev4
    • 18b-dot119-repair-execution-pack.md rev3
    • existing-birth-code-column-patterns-report.md
  2. Run read-only checks:

    • existing 3 fields absent/present;
    • fn hash baseline;
    • trigger count;
    • birth_registry count;
    • DOT-119 script ls/md5/wc;
    • grep for function replacement;
    • schema/migration DOT tool search;
    • current birth triggers and trigger defs;
    • collection_registry governance roles needed for mapping;
    • information_schema column existence for parsed trigger args.
  3. Produce mapping table:

    • collection_name;
    • trigger_name;
    • function_name;
    • trigger_arg;
    • proposed_strategy;
    • proposed_birth_code_column;
    • proposed_source;
    • confidence;
    • notes/provenance.
  4. Explicit manual rows:

    • information_unitsynthetic_id, source manual, provenance 18a rev4 + birth-code report.
    • unit_versionsubordinate, source manual, provenance 18a rev4.
  5. Decision flags:

    • any schema/migration tool relevant? yes/no;
    • any confidence != high? yes/no;
    • are 3 fields already present? yes/no;
    • is DOT-119 stale v1 confirmed? yes/no;
    • is fn hash baseline captured? yes/no.
  6. Upload P0 report to KB path above.

  7. HARD STOP.

Hard boundaries

  • No DDL.
  • No UPDATE/INSERT/DELETE to runtime DB.
  • No DOT-119 execution.
  • No script rewrite.
  • No function edits.
  • No IU trigger.
  • No IU rows.
  • No Pack 2B.

Current state

18b rev3 is acceptable as the umbrella plan. The next actual deliverable is a P0 read-only prompt, not execution of DDL/seed/script work.

Back to Knowledge Hub knowledge/dev/laws/dieu44-trien-khai/reviews/gpt-review-18b-dot119-repair-execution-pack-rev3-2026-05-05.md