KB-1FE2

12 — Recommended Next Macro

4 min read Revision 1
architecturesecond-passnext-macronext-promptsreuse-first

12 — Recommended Next Macro

The second pass clears the GPT gate: the first pass is independently verified, no missed law/engine, reuse-first holds. Implementation may now begin — starting with the read-only/rehearsal macros, mutation only after human Đ32 where noted.

Primary recommendation — run P1 first, AMENDED

OFFICIAL_LISTS_PIVOT_COVERAGE_AND_COUNT_INTEGRITY_RECONCILE [RO] — the first pass's P1, with one mandatory amendment from this pass:

  • ❌ DROP "re-enable trg_auto_sync_registry_counts" — it is ENABLED (tgenabled='O'); re-enabling is a no-op.
  • ✅ INSTEAD: read-only run test_counting_contract() (T1–T8 harness) to characterize the record_count ↔ recompute(count_b) LỆCH on CAT-006/007/023/ALL/CMP/MOL/MAT; identify whether the cause is (a) refresh_meta_catalog_from_pivot skipping non-total CATs, (b) divergent predicates, or (c) never-refreshed CATs (CAT-MAT record 0 vs 55).
  • ✅ RECONCILE CAT-006 (309) vs CAT-DOT (307) → propose the single canonical CAT for dot_tools.
  • ✅ Draft (no apply) the pivot_definitions INSERTs for the 4 lists lacking a pivot (normative_registry, IU/edges, users/roles, event_type_registry).
  • ✅ Fix-spec for PIV-104 (publishedactive).
  • Output: count-integrity remediation packet for human apply. No mutation.

Why P1 first

Correct counting underpins every other official list. The LỆCH is real and currently mis-diagnosed in the first pass — fixing the diagnosis is the highest-value next step and is fully read-only.

Full sequence (carried from first pass doc 12, gate-cleared)

  1. P1 [RO] — pivot coverage + count-integrity reconcile (AMENDED as above).
  2. P2 [RO/rehearsal] — DOT quality retrofit packet (fill operation 50/309, coverage 103 null, tier 19 null via dot-metadata-fill; PIV-104 fix; Đ35 §11 paired-test stage for 178 unpaired). BEGIN..ROLLBACK only.
  3. P3 [Đ28, Đ32-gated] — UI Preview Master → PG-backed living list (reconcile ui_pages 37 / manifest 61 / CAT-007 52; EXTEND ui_pages vs birth ui_surface; NEW surface pivot). KB-only fix now: correct stale review-log.md 19/4 → 61/16.
  4. P5 [Đ38, Đ32-gated] — register Đ23 + Đ45 into normative_registry; deprecate/realign law_catalog (fix dieu28 mislabel); add normative_registry count pivot. No self-approval; ≥2 cross-sign.
  5. P4 [Đ7/37, Đ32-gated] — MOWD workflow-design IU-binding EXTEND + NEW workflow_trigger_design.
  6. P6 [Đ32, human] — candidate-registry field/form/tier activation (field_registry = overlay over directus_fields, not copy).
  7. P7/P8 [RO, parallel] — event-route activation readiness; list IU/KG semantic enrichment.

P1 → P2 → P3 → P5 → P4 → P6 → P7/P8. P1 amended is the single most important next action. The user's core concern ("we may have forgotten an existing capability") is now answered: we have not — the second pass proves the engines exist and are live; the work is reconcile + extend + reuse, not build.

One-line for the conductor

Every musician is identified and present. The only correction to the first score: the metronome (trg_auto_sync_registry_counts) was never off — the drift is in how two sections count the same bar (record_count vs recompute, CAT-006 vs CAT-DOT). Tune that, then proceed.

Back to Knowledge Hub knowledge/dev/reports/architecture/law-capability-discovery-second-pass-cross-check-2026-05-30/12-recommended-next-macro.md