KB-7C76

P3D Pack 1 Phase 4B — Species/Composition Design & Discovery Prompt Report

4 min read Revision 1
p3dpack1phase4breportdesigndiscovery

P3D Pack 1 Phase 4B — Species/Composition Design & Discovery Prompt Report

Date: 2026-05-11 Author: Opus 4.7 Directive: gpt-directive-opus-p3d-pack1-phase4b-species-composition-registry-alignment-2026-05-11.md


1. Status flags

phase_status=PASS
mode=DESIGN_AND_READONLY_PROMPT_ONLY
phase4_vocab_status=PASS_ACCEPTED
migration_allowed=false
species_seed_allowed=false
composition_seed_allowed=false
no_hardcode_contract=included
read_only_prompt_created=true
requires_GPT_User_review_before_dispatch=true

2. Output paths

Artifact Path
Design knowledge/dev/laws/dieu44-trien-khai/design/p3d-pack1-phase4b-species-composition-registry-alignment-design.md
Discovery prompt knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase4b-readonly-species-composition-registry-discovery-prompt.md
This report knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase4b-species-composition-design-and-discovery-prompt-report.md

3. Top 7 design questions/decisions for GPT/User

  1. Species for IU: new or existing? Design recommends new species (E1). None of the 40 existing species fits "information unit." But should there be one species for all IU or per-unit_kind (SPE-IUL for law_unit, SPE-IUD for design_doc_section)?

  2. Composition level: atom, molecule, or compound? Design recommends atom (F1). A law_unit is a self-contained text block; its sections are sub-atomic field values, not separate entities. Per Điều 0-B: no-entity-inside = atom.

  3. Birth registry backfill or forward-only? Design recommends backfill + forward (G1). 12 existing IU birth_registry rows have NULL species/composition. Backfilling them avoids "before/after" split.

  4. Birth gate upgrade or passive registry alignment? Design recommends passive (keep birth gate Tier-0, rely on fn_birth_registry_auto + species_collection_map). Birth gate upgrade = function patch, separate governance decision.

  5. Parent-child during migration: populate or defer? Design recommends populate parent_or_container_ref during migration (H1), defer universal_edges to DOT enrichment (H2). Keeps migration narrow while preserving structural hierarchy.

  6. Collection registry promotion: observed→governed? Must decide before migration. If promoted, all governance requirements (species, labels, etc.) apply. If kept observed, IU stays pilot. Discovery will show current state and the pattern for governed collections.

  7. unit_version strategy: subordinate confirmed? Phase 3 found collection_registry.birth_code_strategy='subordinate' for UV. Does this mean UV should NOT have its own species_collection_map entry? Discovery will check fn_birth_registry_auto logic to confirm.

4. Discovery prompt structure

8 blocks, all read-only, introspect-first:

  • D0: Schema introspection (9 sub-blocks for 6 tables + 3 constraint sets)
  • D1: Full entity_species catalog (columns discovered in D0, not assumed)
  • D2: Species-collection mapping landscape + IU/UV gap
  • D3: Collection registry IU/UV/TAC state
  • D4: Birth registry species/composition for IU/UV
  • D5: Labels + edges coverage + governed exemplar pattern
  • D6: fn_birth_registry_auto source inspection
  • D7: Candidate analysis (fuzzy, labelled candidate_not_approved)

5. Confirmation

GPT/User review is required before dispatching the discovery prompt. No agent, no seed, no migration at this stage.

phase4b_design=CREATED
phase4b_discovery_prompt=CREATED
agent_dispatch_allowed=false_until_GPT_USER_REVIEW
next_action=GPT_REVIEW_PHASE4B_DESIGN_AND_DISCOVERY_PROMPT
Back to Knowledge Hub knowledge/dev/laws/dieu44-trien-khai/reports/p3d-pack1-phase4b-species-composition-design-and-discovery-prompt-report.md