GPT Review — P3D Pack 1 Phase 4 Readiness Deliverables + No-Hardcode Patch

Date: 2026-05-11 Reviewer: GPT-5.5 Thinking / Incomex Hội đồng AI Reviewed:

• knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase4-governed-migration-readiness-no-hardcode-design.md
• knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase4-readonly-governance-vocab-discovery-prompt.md
• knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase4-governance-vocab-species-prep-implementation-prompt-DRAFT.md
• knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase4-readiness-design-report.md

Verdict

Design direction accepted. Discovery prompt rev1 requires patch. Implementation DRAFT remains blocked.

Opus correctly shifted Phase 4 to governed migration readiness and no-hardcode discipline. However, the deliverables still contain small but important hardcode leaks and runtime assumptions that must be corrected before dispatch.

Accepted

1. Phase 4 must be discovery-first.
2. No migration is allowed yet.
3. Hash policy should likely be IU recompute + TAC source hash/provenance, but final implementation must use a real target field/table proven by schema/design.
4. Vocab seed must be generated from live discovery deltas, not typed from memory.
5. Species/composition/governance cannot be assumed complete.
6. Parent-child depth must be discovered, not assumed.
7. Hardcode is now a formal blocker.

Required corrections

1. Avoid “2 rows” as an implementation assumption

Design says “seed species_collection_map for IU/UV — 2 rows”. This may be directionally true for the current two native collections, but it must be written as:

seed rows for discovered target collections that require species mapping

not a fixed number.

2. Avoid “1 doc (~23–36 rows)” as a fixed pilot assumption

Pilot scope must be selected by live query, not memory. The pilot may be one publication, but its unit count must be discovered at runtime.

3. content_profile.source_hash is not proven as an IU target

Design recommends storing TAC source hash in content_profile. That is acceptable only if the target field exists and is semantically appropriate. Phase 4 discovery must check IU columns/profile options before locking this. Otherwise use a traceability table/design later.

4. Discovery prompt hardcodes container/db/user

Patch required to use PG_CONTAINER, PG_DB, PG_USER environment variables and set read-only session.

5. Discovery prompt assumes universal_edges columns

Prior reports showed possible names such as source_ref/target_ref, while prompt used source_entity/target_entity. Prompt must introspect columns before edge queries.

6. Implementation DRAFT remains blocked

No seed/mutation prompt may be dispatched until discovery report is reviewed and GPT/User explicitly approve concrete delta-generation rules.

Decision

GPT will patch the read-only discovery prompt to rev2 and approve it for agent dispatch. Opus does not need to rewrite all design documents now, but future design/prompt language must follow this corrected no-hardcode contract.

Status

phase4_design=ACCEPTED_WITH_NO_HARDCODE_CORRECTIONS
phase4_discovery_prompt_rev1=PATCH_REQUIRED
phase4_discovery_prompt_rev2=TO_BE_PATCHED_BY_GPT
implementation_draft=BLOCKED_NOT_FOR_DISPATCH
migration_allowed=false