P3D Pack 1 Phase 4C — Read-Only / Dry-Run Species Mapping + QT-001 Report

Date: 2026-05-11 Mode: READ-ONLY / DRY-RUN — no writes, no seeds, no backfill Executor: Opus 4.7 (1M) Prompt: knowledge/dev/laws/dieu44-trien-khai/prompts/p3d-pack1-phase4c-readonly-dryrun-species-mapping-qt001-prompt.md (rev3, GPT-approved 2026-05-11) RUN_MARKER: p3d-pack1-phase4c-dryrun-2026-05-11-opus47

0. SCOPE BLOCK (echoed from prompt §0)

target_collection_primary       = 'information_unit'
target_collection_subordinate   = 'unit_version'
target_unit_kind_committed      = 'law_unit'
discriminator_candidate_column  = 'unit_kind'
target_vocab_key                = 'vocab.unit_kind.law_unit'

No other literal decision values used as selectors. Semantic field labels (species_code, composition_level, management_mode, governance_role, birth_code_strategy) appear only as candidate column names verified via information_schema.columns.

1. GATE-0 — Scope preflight (live PG)

Check	What	Live result	Status
P1	`pg_class` row for `information_unit` (relkind='r')	count=1	PASS
P2	`pg_class` row for `unit_version` (relkind='r')	count=2 (`public.unit_version` + `sandbox_tac.unit_version`); public present	PASS (with note: a second `unit_version` table exists in schema `sandbox_tac`; out-of-scope, flagged)
P3	`collection_registry` row for `information_unit`	count=1	PASS
P4	`collection_registry` row for `unit_version`	count=1	PASS
P5	Column `unit_kind` exists on `information_unit`	count=1	PASS
P6	`dot_config` key=`vocab.unit_kind.law_unit`	count=1, value=`law_unit`	PASS

gate0_all_passed = true. No goal BLOCKED by GATE-0.

2. Column introspection per table

2.1 `public.species_collection_map`

Columns present: id, sort, user_created, date_created, user_updated, date_updated, species_code, collection_name, is_primary, discriminator_field, discriminator_value, discriminator_operator, discriminator_config.

All 4 discriminator-candidate columns present. No FIELD_ABSENT.

2.2 `public.entity_species`

Columns present: id, sort, user_created, date_created, user_updated, date_updated, code, species_code, display_name, composition_level, management_mode, prefix, description, status, _dot_origin, parent_id, depth, kg_metadata.

Semantic field presence map:
- IDENTITY: id, code, species_code, display_name, prefix
- CLASSIFICATION: composition_level, management_mode, status
- HIERARCHY: parent_id, depth
- OTHER: description, _dot_origin, kg_metadata, sort, user_*, date_*
No FIELD_ABSENT for required comparison columns.

2.3 `public.collection_registry`

Columns present (relevant subset): collection_name, governance_role, storage_role, source_kind, migration_state, species_code, birth_code_strategy, birth_code_column, birth_identity_source, classification, code, name, name_en, description, purpose, _dot_origin, group, icon, field_count, has_note, description_policy, owner.

No FIELD_ABSENT. governance_role and birth_code_strategy both present.

2.4 `public.birth_registry`

Columns present: id, sort, user_created, date_created, user_updated, date_updated, entity_code, collection_name, species_code, composition_level, dot_origin, born_at, governance_role, inspect_pen, inspect_stamp, inspect_gate, certified, certified_at, status.

No FIELD_ABSENT.

2.5 `public.tac_logical_unit`

Columns present: id, canonical_address, doc_code, parent_id, sort_order, section_type, section_code, owner, identity_profile, tier, lifecycle_status, created_at, updated_at.

unit_kind column: FIELD_ABSENT. Closest categorical alternatives observed: section_type (12 distinct values), tier (4 distinct, incl. NULL), lifecycle_status.

3. GOAL RESULTS

G1 — Discriminator usage in production

Discriminator column	Exists	Non-NULL rows
`discriminator_field`	yes	0
`discriminator_value`	yes	0
`discriminator_operator`	yes	0
`discriminator_config`	yes	0

Total species_collection_map rows = 153. Rows with ANY discriminator field set = 0. Sample of non-NULL rows = empty.

Conclusion: discriminator_schema_support = true (columns exist). discriminator_production_active = false (zero rows populated).

G2 — `fn_birth_registry_auto` discriminator handling

Function exists (count=1). Source captured. Species/composition resolution clause:

SELECT scm.species_code, es.composition_level
  INTO v_species_code, v_comp_level
  FROM species_collection_map scm
  LEFT JOIN entity_species es ON es.species_code = scm.species_code
  WHERE scm.collection_name = TG_TABLE_NAME
    AND scm.is_primary = true
  LIMIT 1;

Function source contains no reference to discriminator_field, discriminator_value, discriminator_operator, or discriminator_config. Also no reference to NEW.unit_kind or v_row_json->>'unit_kind'.

Record-field access observed: v_row_json->>v_code_field (trigger arg), v_row_json->>'_dot_origin', v_row_json->>'id'.

Conclusion: uses_discriminators = false. reads_record_fields_beyond_code = true (reads _dot_origin and id, but not any discriminator). Discriminator path is dormant at birth time.

G3 — Complete species landscape

SELECT count(*) FROM entity_species = 40.

Categorical distinct-value counts (live):

composition_level: atom=20, molecule=11, compound=8, meta=1.
management_mode: governed=26, observed=8, excluded=6.
status: active=40.
depth: 1=40 (flat tree).

Full row dump (IDENTITY + CLASSIFICATION + HIERARCHY columns) — 40 rows: see live PG (excerpt at execution time).

Highlights of the 8 observed species (relevant to G8):

species_code	code	display_name	composition_level	management_mode
ai_support	SPE-AIS	Hỗ trợ AI	atom	observed
business_support	SPE-BZS	Hỗ trợ Kinh doanh	molecule	observed
checkpoint_support	SPE-CKS	Hỗ trợ Checkpoint	molecule	observed
directus_field	SPE-DXF	Directus Field	atom	observed
governance_infra	SPE-GOV	Hạ tầng Giám sát	atom	observed
help_center	SPE-HLP	Trung tâm Trợ giúp	molecule	observed
pivot_result	SPE-PVR	Ket qua Pivot	molecule	observed
system_log	SPE-LOG	Nhật ký Hệ thống	atom	observed

G4 — Collection-registry context for targets

field	information_unit	unit_version
id	184	185
code	COL-176	COL-177
governance_role	observed	observed
storage_role	primary	primary
source_kind	native	native
migration_state	pilot	pilot
species_code	NULL	NULL
birth_code_strategy	synthetic_id	subordinate
birth_code_column	NULL	NULL
birth_identity_source	manual	manual
classification	NULL	NULL

Both targets: governance_role=observed, no species assigned.

G5 — Birth-registry state for targets

collection	total rows	species_code NULL	composition_level NULL
information_unit	12	12	12
unit_version	0	0	0

First 12 information_unit rows: all entity_code=information_unit::<uuid>, species_code=NULL, composition_level=NULL, governance_role=observed, certified=false, status=born. (Full entity_codes captured in run.)

G6 — Discriminator candidate verification on target

information_unit.unit_kind exists (P5 PASS). Distinct values + counts:

unit_kind	count
design_doc_section	12

Total information_unit rows = 12. Only one unit_kind value populated today. The target value law_unit (from scope §0 / dot_config vocab.unit_kind.law_unit) is not yet present in any row.

G7 — Collections sharing exact `birth_code_strategy` with subordinate target

v_sub_strategy = collection_registry.birth_code_strategy WHERE collection_name='unit_version' → subordinate.
Other collections with EXACT birth_code_strategy = 'subordinate': count = 1 (only unit_version itself).
Of those, with species_collection_map rows: 0.

Conclusion: unit_version is the only collection in the entire registry using the subordinate birth strategy. No peer pattern to cross-reference. No fuzzy expansion attempted.

G8 — Dry-run candidate species with deterministic labels

Comparison rule (per prompt §G8): label PLAUSIBLE iff entity_species.management_mode = collection_registry.governance_role for information_unit, both non-NULL; CONFLICT iff both non-NULL and unequal; UNKNOWN if either NULL; N/A if field absent.

Target gov_role for information_unit = observed. Both columns present → no N/A or UNKNOWN cases.

Counts:

Label	Count
PLAUSIBLE	8
CONFLICT	32
UNKNOWN	0
N/A	0

PLAUSIBLE candidates (each row labelled candidate_not_approved):

species_code	proposed_composition_level	compatibility_reason
ai_support	atom	management_mode=observed matches target governance_role=observed
business_support	molecule	same
checkpoint_support	molecule	same
directus_field	atom	same
governance_infra	atom	same
help_center	molecule	same
pivot_result	molecule	same
system_log	atom	same

CONFLICT rows (32) include all governed species (e.g. law, jurisdiction, law_enforcement, dot_tool, workflow, …) and all excluded species. Reason: management_mode (governed or excluded) ≠ target governance_role (observed).

All G8 output rows carry candidate_not_approved.

G9 — QT-001 backfill dry-run per PLAUSIBLE candidate

Population of NULL species rows: 12 (information_unit, all current rows).

proposed_species	proposed_composition_level	rows_would_update
ai_support	atom	12
business_support	molecule	12
checkpoint_support	molecule	12
directus_field	atom	12
governance_infra	atom	12
help_center	molecule	12
pivot_result	molecule	12
system_log	atom	12

Sample (first 3 rows × 8 candidates = 24 projection rows captured), e.g.:

proposed_species	proposed_comp	entity_code	current_species	current_comp	state
ai_support	atom	information_unit::3ffbbaa5-…	NULL	NULL	candidate_not_approved
ai_support	atom	information_unit::60c0017d-…	NULL	NULL	candidate_not_approved
ai_support	atom	information_unit::8e247041-…	NULL	NULL	candidate_not_approved
governance_infra	atom	information_unit::3ffbbaa5-…	NULL	NULL	candidate_not_approved
…	…	…	…	…	…

Every output row: candidate_not_approved. No INSERT/UPDATE performed.

G10 — TAC unit_kind distribution snapshot

information_schema.columns lookup for tac_logical_unit.unit_kind → 0 (FIELD_ABSENT).

Closest categorical column observed: section_type (12 distinct values, total 86 rows).

section_type	count
technical_spec	19
governance_process	16
paragraph	14
heading	10
principle	8
process	7
checklist	6
changelog	2
article	1
appendix	1
instruction_block	1
definition	1

tier distribution: NULL=59, unit=22, section=4, root=1. None equals law_unit.

No exact value in tac_logical_unit matches target_unit_kind_committed='law_unit' via either section_type or tier. TAC data cannot directly inform unit_kind discriminator strategy on information_unit without a separate mapping step. Reported as FIELD_ABSENT on the original lookup; no fuzzy expansion attempted.

4. PHASE 4B SNAPSHOT RE-VERIFICATION

Phase 4B finding	Re-verified live?	Holds?
`species_collection_map` total = 153 rows	yes	✅
0 IU/UV rows in species_collection_map	yes	✅ (only `unit_version` peers checked, 0 maps)
`fn_birth_registry_auto` resolves species via `is_primary=true LIMIT 1`, no discriminator	yes (full source captured)	✅
`entity_species` = 40 species	yes	✅
`SPE-GOV` = governance_infra/atom/observed	yes	✅
`SPE-LAW` = law/compound/governed	yes	✅
12 IU birth rows, all NULL species/composition	yes	✅
0 UV birth rows	yes	✅
Species tree flat (all depth=1)	yes	✅
TAC `parent_id` containment	yes (`parent_id` column exists)	✅

No Phase 4B label was carried over as fact without live re-query.

5. SCALE BRITTLENESS SELF-AUDIT (§6 of prompt)

Check	Answer
Did I list any species/composition/role value from memory?	No — every value queried live.
Did I list any column name as "required" without verifying it exists?	No — every table introspected, or queried via `SELECT ` / generic `count()`.
Did I use any "-like" or semantic fuzzy match for selectors?	No — exact-value match only (`=`, not ILIKE).
Did I assume fixed counts?	No — every count is a live snapshot at run time.
Did I assume fixed field lists for output shapes?	No — projections derived from observed columns; absent fields reported as FIELD_ABSENT.
Would my queries adapt if a column was renamed?	Yes — they introspect column presence before querying and would emit FIELD_ABSENT.
Would my queries adapt if 50 new species were added?	Yes — G3 uses `SELECT count(*)` + `GROUP BY` on categorical columns, no enumeration.
Did I report Phase 4B labels as facts without re-verification?	No — every label re-verified in §4.
Did I apply subjective judgment in G8/G9 labels?	No — purely the rule `management_mode = governance_role`.

scale_brittleness_self_audit_passed = true.

6. CRITICAL OBSERVATIONS (informational, no decision locked)

Discriminator runtime is dormant. Schema supports it (species_collection_map has all 4 columns), but 0 rows populated and fn_birth_registry_auto does not read those columns. Any discriminator-based strategy requires either a function patch (out of scope) or a post-birth correction tool (out of scope).
subordinate peer pattern is unique. unit_version is the only collection with birth_code_strategy='subordinate' registry-wide. There is no peer mapping to copy.
law_unit does not exist as a live value anywhere. Neither information_unit.unit_kind nor tac_logical_unit.section_type/tier contains law_unit. dot_config['vocab.unit_kind.law_unit'] = law_unit is a vocabulary entry; no current row consumes it. Phase 5 TAC migration would be the first event to introduce that value.
8 PLAUSIBLE species, but composition spans atom/molecule (5 atom, 3 molecule). The 12 pilot rows (design_doc_section) have no nesting evidence in information_unit; choosing molecule would imply containment that the data does not currently exhibit. This is a decision for GPT/User — Phase 4C does not lock it.
Schema duplication for unit_version. A sandbox_tac.unit_version table exists alongside public.unit_version. Confirm Phase 5 migration targets public.unit_version specifically; flag separately if sandbox_tac is dead schema.

7. STATUS FLAGS

phase4c_dryrun_status=PASS
no_mutation_performed=true
gate0_all_passed=true
discriminator_runtime_active=false
discriminator_column_exists_on_target=true
species_decision_locked=false
composition_decision_locked=false
plausible_candidate_count=8
conflict_candidate_count=32
unknown_candidate_count=0
scale_brittleness_self_audit_passed=true
next_recommended_action=GPT_User_review_8_PLAUSIBLE_candidates_for_information_unit_species_pick_then_decide_QT001_backfill_scope_and_phase5_nesting