Architecture Review Lifecycle
When architectural reviews should occur, and a standard checklist to run each time. This formalises the ecosystem’s existing “phase-gated, reviewable change” discipline (Principle 10; WIL “each phase ends with an architecture review and does not continue until reviewed”). Evidence-first.
Review triggers
| Trigger | Why review | Depth | Output |
|---|---|---|---|
| New repository | Confirm it’s needed & correctly placed (not a pack/skill/feature) | Light–standard | Placement decision + ADR + registry row |
| Major capability | Ownership, reuse, duplication, layer placement | Standard | Capability page + Reuse Map update |
| Platform restructuring | Layer boundaries, dependency directions | Deep | ADR (e.g. 0005) + updated diagrams |
| Cross-venture reuse | Prevent trapped/duplicated capabilities | Standard | Extraction decision; Reuse Map update |
| Commercialisation | External exposure, multi-tenancy, guardrails | Deep | Risk review + ADR |
| New/changed AI autonomy | Authority boundary + approvals | Standard | AI Governance entry + ADR |
| Annual strategic review | Whole-ecosystem health, drift, roadmap | Deep | Refreshed Architecture Review + risks |
| Phase boundary (portal) | Gate before the next phase | Standard | Phase changelog + review notes |
Review depth
graph LR
T[Trigger] --> D{Structural or cross-repo?}
D -->|no| L[Light review<br/>Definition of Done + consistency]
D -->|yes| S{Changes layer model / dependencies / autonomy?}
S -->|no| ST[Standard review<br/>checklist + ADR]
S -->|yes| DP[Deep review<br/>full checklist + risks + evolution impact]
Standard review checklist
Run this checklist at every review; deep reviews add the starred (★) items.
Placement & necessity
- [ ] Passes the four-part test (real problem · reduces coupling · clear owner · one-paragraph)
- [ ] Correct destination per the Decision Matrix
- [ ] Not a reusable capability being trapped in a venture repo
- [ ] Not duplicating an existing concept (Concept Ownership Registry)
Ownership & dependencies
- [ ] Single, clear owner named (no diffuse ownership)
- [ ] Dependencies point at contracts, not venture repos; no new circular/hidden dependency
- [ ] ★ Impact on the system dependency map assessed
Evidence & documentation
- [ ] Claims cite evidence; uncertainty recorded (Principle 8)
- [ ] Uses the canonical template; frontmatter correct
- [ ] Digest / capability page / diagrams updated; registry row present
- [ ]
llms.txt+ current-state (changelog, dashboard) updated
Decision & governance
- [ ] Significant decision recorded as an ADR (proposed → accepted with human sign-off)
- [ ] Required human approvals identified (AI Governance)
- [ ] ★ Reuse & duplication re-checked (Capability Reuse Map)
- [ ] ★ New risks logged (Architecture Risks)
Consistency (Definition of Done)
- [ ] Registry ⇄ digests ⇄ diagrams agree; no orphans, no dead links
- [ ]
last_revieweddates current
Cadence
- Triggered reviews — on any trigger above (event-driven).
- Annual strategic review — a scheduled deep review refreshing the Architecture Review, Risks, Opportunities, and Evolution pages.
- Portal phase gates — each phase (A–F and beyond) ends with a review before the next begins.
Implemented vs recommended
- ✅ Implemented: phase-gated reviews (portal phases A–F; WIL phase gates); the Phase-D ecosystem review; ADR sign-off.
- ⏳ Recommended: a standing annual review cadence; automated pre-review checks (link/orphan/pin-freshness) via the orchestrator so human review focuses on judgement, not mechanics.