Incomex AI Portal
KB-6AD6

GPT Review — Master Design Rev3 Gaps MP-D16..MP-D22 (2026-05-28)

7 min read Revision 1
reviewgptmaster-design-rev3mp-d11mp-d15governancekaizenobservabilitymp-d16mp-d222026-05-28

GPT Review — Master Design Rev2 Revision 3 MP-D11..MP-D15

Date: 2026-05-28 Reviewer: GPT Council via Web Connector fallback

Inputs read directly

KB:

  • knowledge/dev/design/v0.6-iu-4mothers-event-foundation-rev2/00-master-design-rev2.md revision 3
  • knowledge/dev/design/v0.6-iu-4mothers-event-foundation-rev2/07-master-design-rev2-report.md revision 3
  • knowledge/dev/design/v0.6-iu-4mothers-event-foundation-rev2/08-bidirectional-input-kaizen-governance-addendum.md revision 1

Uploaded files searched/read via file search:

  • Bắt sự kiện của PG.docx uploaded 2026-05-28
  • earlier Bắt sự kiện của PG.docx uploaded 2026-05-27
  • 4 mẹ mở rộng.txt

Verdict

REV3_STRONG_BUT_NEEDS_SECOND_ORDER_GOVERNANCE_PATCH

MP-D11..MP-D15 are applied and materially improve the design. The revision now captures:

  • bidirectional input flow;
  • staging vs PG canonical boundary;
  • simple Kaizen intake;
  • T6→T1 MOW hierarchy canvas;
  • MOT/JFT matrix;
  • four separate UI surfaces;
  • governance cockpit and AI/Agent ops console;
  • reuse-first and law/no-double-ownership matrix.

However, before design approval for downstream implementation, the design should receive a second-order governance/operability patch. The patch is not a rewrite. It should close several gaps that become important at high scale.

Confirmed PASS items

  1. MP-D11 PASS: Read and write data planes are distinct; staging is not SoT; PG canonical remains SoT.
  2. MP-D12 PASS: Unified MOW canvas T6→T1 is explicitly documented.
  3. MP-D13 PASS: Kaizen is kept simple; ordinary users do not edit workflow logic.
  4. MP-D14 PASS: MOT/JFT envelope matrix is explicitly documented.
  5. MP-D15 PASS: Personal Dashboard, MOW Canvas, Governance Cockpit, and AI/Agent Ops Console are separated.
  6. Event 5-layer remains compatible with the uploaded PG event document.
  7. Assembly First is preserved: OSS tools are candidate adapter slots only.
  8. No production mutation reported.

Second-order gaps requiring patch

MP-D16 — Connect T6→T1 operating hierarchy to Điều 5 five-tier architecture

The current design adds T6→T1 business hierarchy, but it does not explicitly map it to the older 5-tier architecture law:

  • Tầng 1 Hạ tầng;
  • Tầng 2 Cơ sở / registries / metadata / DOT / fields / taxonomy;
  • Tầng 3 Modules nền tảng;
  • Tầng 4 Chuyên môn / nghiệp vụ;
  • Tầng 5 Giám sát + cải tiến.

The design must clarify that T6→T1 is the business operating hierarchy inside Tầng 4/5, while fields/registries/DOT belong to Tầng 2, modules/MOW/MOT/MOIT/MOUT belong to Tầng 3, governance/Kaizen cockpit belongs to Tầng 5. This prevents confusing the two tier systems.

MP-D17 — Governance cockpit queue taxonomy and lifecycle

The Rev3 cockpit lists panels, but still needs a formal queue taxonomy and lifecycle suitable for 20,000+ items:

  • problem classes;
  • severity levels;
  • ownership/assignment;
  • triage state;
  • SLA/SLO state;
  • escalation path;
  • dedupe/suppression/grouping;
  • acknowledgement vs resolution;
  • snooze/waive with approval;
  • incident/problem/change separation.

Without this, the cockpit is still a set of panels, not yet an operations governance model.

MP-D18 — Kaizen intake anti-noise and duplicate-control lifecycle

Rev3 includes Kaizen metrics, but it should also design the lifecycle:

  • received;
  • auto-classified;
  • duplicate_suspected;
  • needs_clarification;
  • accepted_for_review;
  • rejected_noise;
  • converted_to_proposal;
  • approved;
  • merged;
  • measured_after_change.

Need duplicate detection and queue controls so Kaizen does not become noise at company scale.

MP-D19 — Input canonicalization policy and data-quality guardrails

Direct/no-op canonicalization is correct, but needs guardrails:

  • what kinds are allowed to bypass workflow;
  • required validation level;
  • correction model after direct write;
  • data lineage from staging to canonical;
  • retention/PII policy for staging text/audio/attachments;
  • audio transcription confidence and human-review thresholds;
  • idempotency for direct canonicalization;
  • abuse/spam/rate limit controls.

MP-D20 — Standard-industry observability profile should be clearer

Uploaded PG event document emphasizes schema registry, distributed tracing, workflow orchestrators, and governance UI. Rev3 mentions these, but should add a compact Minimum Observability Profile:

  • schema registry / event validation;
  • trace_id + correlation_id;
  • event lag p50/p95/p99;
  • worker heartbeat and silent worker;
  • DLQ depth and replay outcomes;
  • idempotency conflicts;
  • queue lease timeout;
  • governance summary freshness;
  • AI/Agent task status;
  • top blocked/cannot_complete clusters.

Also define what remains machine-only vs human-visible.

MP-D21 — OSS candidate strategy update after raw uploaded docs

Rev3 references candidate OSS slots, but a small addendum should explicitly reconcile the raw uploaded document's suggestions:

  • Hasura: sandbox/reference_only or reject_as_core_owner; useful to study, not core due to Directus/Nuxt/PG boundary and no hidden SoT.
  • pg-boss/Graphile Worker: preserve future_adapter_slot only if adapter maps states and does not own state vocabulary; otherwise reject as primary substrate now.
  • Benthos/Redpanda Connect: future CDC adapter for external mirroring/high-volume table capture; not domain event SoT.
  • NATS/Redis Streams: future transport; not SoT.
  • Temporal/Camunda/Airflow: future execution-backend/reference slots only; MOW/PG registry remains workflow definition SoT.
  • Watermill: reference-only or future glue library; no immediate dependency.
  • OpenTelemetry/Jaeger: future observability adapters; trace shape already compatible.

MP-D22 — Survey sequencing should include Tier Registry Survey and Governance Ops Survey

Rev3 recommends Candidate Registry Survey and Tier Registry Survey. Add a third survey macro:

IU_4MOTHERS_GOVERNANCE_OPS_SURVEY_DOCUMENT_ONLY_*X

Survey existing:

  • task status queues;
  • AI task queue/agent run tables;
  • existing Directus task collections;
  • existing governance views/logs;
  • existing worker heartbeat data;
  • existing event/problem categories;
  • whether any existing dashboard/ops module can be reused.

Do not proceed to implementation yet.

Run a document-only patch macro for MP-D16..MP-D22:

IU_4MOTHERS_MASTER_DESIGN_GOVERNANCE_SECOND_ORDER_PATCH_DOCUMENT_ONLY_3000X

Then GPT should read revision 4 directly.

Notes

This patch is deliberately second-order. Rev3 is not wrong; it is now broad enough that the next risk is operational ambiguity at scale. The goal is to harden governance, noise control, observability, and tier/law mapping before implementation begins.

Back to Knowledge Hub knowledge/dev/reports/architecture/iu-4mothers-master-design-rev3-gpt-review-gaps-mp-d16-d20-2026-05-28.md