Trigger: Rich audit directive 2026-05-01T18:30 BST — “audit the design spec itself for whether ζ-Q15-Q18 should have been deferred to Phase 4-5. if a question has been asked prematurely, then do not try to preserve its answer: we should just discard it and go back to the point where we deviated away from the optimum course.”
Audit verdict per design spec v1.1 §10 phase definitions:
| Q | A-num | Topic | Asked at | Correct phase | Phase delta |
|---|---|---|---|---|---|
| ζ-Q15 (REVISED) | A-142 | AI agent stack revisit (Π.θ ★ L4M+ASDK+Skills+AGT) | Phase 3 | Phase 5 (IAS+IK §8 open Q’s) | +2 phases too early |
| ζ-Q16 (REVISED) | A-143 | Will Update Architecture B2B+B2C (Σ.κ ★) | Phase 3 | Phase 6 (per-brand integration — IW) | +3 phases too early |
| ζ-Q16 (REVISED-AGAIN) | A-144 | Σ.μ.ε.5 ★ IW productisation correction | Phase 3 | Phase 6 (per-brand integration — IW) | +3 phases too early |
| ζ-Q17 | A-145 | Trigger-event detection (Λ.η ★ Tiered) | Phase 3 | Phase 6 (per-brand integration — IW + MFI) | +3 phases too early |
| ζ-Q18 | A-146 | InheritKitBench v0.1 commitment (Ω.η ★) | Phase 3 | Phase 5 (AI-vendor commercial / closure prep) | +2 phases too early |
Deviation point: 2026-05-01T09:35 BST when ζ-Q15 was reframed as AI agent stack revisit (Phase-5 scope) instead of either (a) original t-box layering R2-completeness check (correctly deferred to Phase 5) or (b) Wills follow-through (correct Phase 3 next-step per A-124).
Lesson — codification candidate feedback_phase_fit_audit_2026_05_01: Spikes can run when evidence surfaces (preserves “spikes must not be delayed” discipline) BUT the resulting Q must be queued at the correct phase per design spec v1.1 §10, NOT promoted to next-Q just because the spike was good. Decoupling spike-evidence-gathering from next-Q-phase-placement is the discipline.
Files reverted (state file changelogs preserved as historical-only):
| File | From | To |
|---|---|---|
inherit-v2-architecture-state.md | v3.32 | v3.27 (logical equivalent) |
inheritkit-architecture-state.md | v1.6 | v1.1 |
ias-architecture-state.md | v1.7 | v1.2 |
2026-04-29-multi-phase-audit/INDEX.md | v2.22 | v2.17 |
inherit-v2-risk-register.md | v1.3 | v1.2 (risk count 45→34) |
inherit-v2-standards-engagement-state.md | v1.1 | v1.0 |
richard-tasks-tt.md (docs-personal) | v1.20 | v1.15 (tasks #205..#224 removed) |
Cascade Q files moved:
Q-015-zeta-ai-agent-stack-revisit.md→_deferred-questions/Q-015-zeta-ai-agent-stack-revisit-DEFERRED-to-phase-5.mdQ-016-zeta-will-update-architecture.md→_deferred-questions/Q-016-zeta-will-update-architecture-DEFERRED-to-phase-6.mdQ-017-zeta-trigger-event-detection-living-monitoring.md→_deferred-questions/Q-017-zeta-trigger-event-detection-DEFERRED-to-phase-6.mdQ-018-zeta-inheritkitbench-v0-1-commitment.md→_deferred-questions/Q-018-zeta-inheritkitbench-DEFERRED-to-phase-5.md
Spike files: status locked-outcome → research-only-deferred. Library research preserved + pre-loaded for re-use when correct phase opens.
What survives unchanged:
- Phase 2 closure (ζ-Q1..ζ-Q8 + MQ-000 + MQ-001 R2 + MQ-002) — all locks preserved
- Phase 3 derivative-Q’s (ζ-Q9..ζ-Q14 / A-136..A-141) — all schema/primitive Phase-3-correct
- Master plan v1.10 unchanged
- Per-repo BUILD-PLAN.md v1.4-1.6 unchanged
- pin_kind tagging convention (separate work earlier today; not part of cascade)
- 2026-05-01-inheritkitbench-explainer.md (general explainer; not a lock)
- Susskind 2008 + 2023 library indexing (always useful)
Phase 3 status post-revert: 6 Q answered / 30-45 budget / 13-20% complete. NEXT = ζ-Q15 first Wills follow-through Q per A-124 module-authoring sequence (Will + Bequest + SubstitutionClause schema authoring across 8 jurisdictions).
Standard end-of-turn directive refined post-reversal — Rich’s standard text updated 2026-05-01T18:45 with explicit phase-fit hard guard + Phase-N forbidden-Q list + spike-decoupling-from-next-Q discipline.