GPT Review — P3D Pack 1 Phase 5B rev2 Split OK; 5C Drafts Need rev1 Prep

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

• knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase5b-hybrid-nesting-species-pilot-migration-design.md rev2
• knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase5c1-species-mapping-qt001-backfill-prompt-DRAFT.md
• knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase5c2-dieu35-hybrid-pilot-migration-prompt-DRAFT.md
• knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase5b-rev2-design-and-split-prompts-report.md

Verdict

Phase 5B rev2 split is accepted directionally.

The split into 5C1 and 5C2 is correct and materially safer than the original all-in-one draft.

Neither 5C1 nor 5C2 is dispatch-approved yet.

5C1 is now the immediate next focus because 5C2 depends on 5C1 completion.

What is accepted

1. 5C1 = QT-005 species/mapping prep + QT-001 backfill only.
2. 5C2 = DIEU-35 pilot migration only.
3. D3a hybrid remains locked for pilot.
4. parent_or_container_ref=NULL remains locked for pilot.
5. Composition for pilot = atom remains locked.
6. unit_version species mapping = no for now.
7. Governance role stays observed.
8. Rollback capture target = KB/report artifact + VPS log, no new control table.
9. 5C2 is blocked on 5C1 completion and publication_authority_ref resolution.
10. Historical counts are now treated as evidence, not executable gates.

Remaining issues in 5C1 DRAFT

1. Species identity is still unresolved

5C1 cannot become rev1 until GPT/User locks the species identity fields.

Opus left eight placeholders. That is correct for DRAFT, but the next step must turn these into a decision memo rather than guessing inside the execution prompt.

2. QT-005 insert needs a schema-driven fill policy

5C1 says Agent builds INSERT from introspected NOT NULL columns. This is directionally correct, but dispatch-safe rev1 must also define what to do with each introspected required column:

provided_by_GPT
provided_by_default
provided_by_existing_db_default
FIELD_UNRESOLVED_STOP

Without this matrix, Agent may improvise values for unexpected NOT NULL columns.

3. Taxonomy parent selection must be decided before dispatch

The prompt says discover candidate parent in INTROSPECT-4 and use GPT/User locked parent if provided. Good, but rev1 must not let Agent choose the parent. Parent id/code must be locked or the run must stop.

4. The “species_code” vs “entity_code” distinction must be verified

5C1 placeholders include both species_code and entity_code/prefix. Prior reports show entity_species fields differ from assumed names. Rev1 must avoid assuming the exact semantic relationship between species_code, entity_code, prefix, etc. It should use introspection and a GPT-approved field mapping.

5. Backfill target selection must be exact-key captured before UPDATE

5C1 has this directionally, but rev1 should explicitly split:

SELECT target keys INTO captured list
VERIFY captured list count = live target count
UPDATE WHERE key IN captured list
VERIFY all captured keys updated

Do not rely only on UPDATE ... RETURNING as the first capture point.

Remaining issues in 5C2 DRAFT

5C2 is directionally sound but must wait.

Known blockers:

1. 5C1 completion and GPT acceptance.
2. publication_authority_ref_value must be locked.
3. fn_iu_create return shape must be verified safely without side effects; if planner is insufficient, do not use a write call outside a transaction boundary.
4. JSON key names are acceptable design constants, but rev1 must declare them explicitly as design constants and verify they do not collide with existing identity_profile semantics.

GPT decision for immediate next work

Do not patch 5C2 now.

Open next:

P3D_PACK1_PHASE5C1_SPECIES_IDENTITY_DECISION_AND_REV1_PREP

Mode:

DECISION_MEMO + 5C1_PROMPT_REV1_PREP_ONLY

No Agent dispatch. No DB write.

Recommended species identity direction

GPT accepts Opus's broad direction:

one new atom species for information_unit collection

But exact code/name/prefix/parent must be selected from live taxonomy evidence, not invented casually.

Working label for memo only:

information_unit_atom

This label is not dispatch-approved until the decision memo verifies the entity_species schema and taxonomy parent strategy.

Status

phase5b_rev2=ACCEPTED_DIRECTIONALLY
phase5c1_draft=NOT_DISPATCH_READY
phase5c2_draft=NOT_DISPATCH_READY
next_focus=PHASE5C1_SPECIES_IDENTITY_DECISION_AND_REV1_PREP
agent_dispatch_allowed=false
migration_allowed=false
seed_allowed=false
backfill_allowed=false