GPT Council Review Addendum — Agent Macro Report + Workload/Runtime Calibration

Date: 2026-05-27 Reviewer: GPT Council via Web Connector fallback Scope: Addendum after user supplied Agent final summary for IU_4MOTHERS_EVENT_FOUNDATION_REQUIREMENT_DESIGN_PASS Related review: knowledge/dev/reports/architecture/iu-4mothers-event-foundation-council-review-gpt-2026-05-27.md

User emphasis acknowledged

The Council review must prioritize:

1. New design must wrap and extend old design, reusing what already exists.
2. Constitution/law compliance must be reviewed clause-by-clause, not by vague compatibility claims.
3. PostgreSQL must be maximized as SSOT and execution/governance substrate wherever possible.
4. All previously agreed details must be rechecked; missing details require patch requests.
5. Workload/runtime targets are strategic signals of agent capability. Not every small task must force 45–60 minutes, but work should be scoped large enough where appropriate. Incomex already achieved >100,000x level; reaching 45–60 minute macro depth requires deliberate task sizing.

Runtime/workload calibration decision

The Agent report cooked for ~17m43s while declaring a 45–60 minute target. This is not treated as a hard failure for the requirement/design macro, because:

• The delivered task was document-first and read-only.
• Six KB documents were created and the macro surfaced real gaps without production mutation.
• The user clarified that smaller tasks may accept lower runtime.

However, future prompts should distinguish between:

• target_depth: expected reasoning/evidence depth.
• target_runtime: expected wall-clock effort only when the task scope is large enough.
• minimum_artifact_quality: must not be lowered even if runtime is shorter.

Recommended wording for future macro prompts:

Prefer 45–60m depth for large architecture/implementation macros. For smaller document-only patches, runtime may be lower, but the agent must still meet explicit evidence, law-matrix, reuse-matrix, PG-maximization, and gap-report requirements. Do not pad time artificially. Increase task scope when a 45–60m macro is desired.

Review verdict remains

ACCEPT_WITH_PATCHES remains unchanged.

Agent output is useful and mostly aligned, but implementation Phase 0 should still wait until a document-only patch closes the review gaps.

Additional patch requirement from user emphasis

The patch macro must explicitly add four verification sections to the design pack:

1. Old Infrastructure Coverage Matrix

   • IU foundation: source/domain/tree axis, compose, split/merge, trigger in/out, SQL link, post-CUT materialization.
   • Queue/event: event_outbox, event_type_registry, job_queue, DLQ, heartbeat, event/job distinction.
   • DOT: command catalog, dot_config, gates, DOT-pair mutation discipline.
   • Cut pipeline: MARK, VERIFY, APPROVE, CUT, COMPLETE, post-CUT autowire gap.
   • Directus/Nuxt: PG→Directus→Nuxt, Directus API/admin, Nuxt render shell only.
   • Vector: no cross-IU vector pollution, no Qdrant write without policy.

2. Constitution and Law Clause Matrix

   • For each relevant law/principle: source, binding requirement, design satisfaction, risk, sentinel test.
   • Must include Hiến pháp v4.6.3, Điều 7 Assembly First, Điều 28 display/Nuxt boundary, Điều 30 regression, Điều 31 integrity, Điều 32 approval, Điều 33 PostgreSQL, Điều 35 DOT governance, Điều 37 governance org, Điều 38/39 IU, Điều 45 queue/event/task orchestration, Điều 34 draft decision path.

3. PG Maximization Map

   • PG as SSOT for event ledger, registry, schema JSON, trace/correlation, idempotency, retry, DLQ, replay, workflow/task/field/form/output registry, DOT catalog, governance state, healthcheck, audit evidence.
   • External tools must attach as adapter/transport/executor/gateway only; no second SoT.

4. Agreed Details Completeness Checklist

   • 4 Mothers, Event 5 layers, Assembly First, business hierarchy, JFT, triggers, MOW UI normal/proposal mode, MOT layout, MOIT, MOUT, Governance UI, scale targets, IU compatibility, OSS candidate boundaries, future law boundary.

OSS critique adjustment required

The patch must replace overly final language with decision-class language:

• reject_as_primary_owner_or_canonical_substrate_now
• defer_until_profile_trigger
• future_adapter_slot_preserved
• not_a_second_SoT

This is especially required for pg-boss, Graphile Worker, Temporal, Camunda, Benthos and NATS.

Next macro recommendation

Run this first:

IU_4MOTHERS_EVENT_FOUNDATION_DESIGN_REVIEW_PATCH_1000X_OR_2000X

Scope: KB document-only patch, no PG/Directus/Qdrant mutation.

After patch review PASS, then authorize:

IU_POST_CUT_AUTOWIRE_SPLIT_MERGE_AND_EVENT_FOUNDATION_PHASE0_6000X

Final addendum decision

Agent capability note is accepted. Future macro scoping should avoid artificial runtime padding while still assigning sufficiently large work for high-depth macro targets.

Implementation remains gated by design patch.