Frozen document. Dates and status refer to its original source. Private resources and unreviewed images are unavailable.
Atlas architecture assessment · 30 September 2026
Status: assessment complete; submission lifecycle design approved. Refactor implementation remains blocked by the linked acceptance gates.
This assessment used the improve-codebase-architecture skill and its codebase-design vocabulary. It scoped recent changes through merged PR #52, read the Atlas domain model and accepted ADR 0005 and ADR 0008, then inspected the durable case-operation and Journey capture paths. The visual before/after report was created at [private location omitted] on the owner's Mac; the durable conclusions are recorded here because temporary reports are not permanent evidence.
| ID | Module and evidence | Finding | Recommendation | Owner choice and mitigation |
|---|
| ARCH-2026-09-30-01 | apps/bff/src/cases.ts, case-recovery-routes.ts, camunda.ts | Acceptance, outbox dispatch, owner receipt, and recovery observation share state but are spread across many lifecycle actions. A change requires tracing distant regions of the module. | Strong: explore a deeper durable case-operation module. Keep ADR 0005's owner revision and reconciliation rules. The interface should be the test surface; the current public integration checks are behavior guards. | Owner approved one lifecycle interface and an incremental submission-first extraction. See the approved design and Extract the durable submission lifecycle behind one case-operation interface. No implementation yet; finding remains open. |
| ARCH-2026-09-30-02 | scripts/serve_journey.py, capture_journey_reviews.py, capture_pull_metrics.py | Each new metric edits the HTTP refresh method and browser fallback wiring, increasing the chance of inconsistent source labels under partial failure. | Worth exploring: concentrate snapshot capture and provenance rules behind one module, leaving HTTP serving and source adapters separate. The deletion test suggests this would preserve locality rather than add a pass-through. | Not selected for the next exploration. Reassess when another data source or failure mode changes this path. |
The assessment found architectural friction, not a verified product defect. The heatmap in the Journey Dashboard shows committed file-change activity by module/day; it does not measure code quality or prove these findings. The next code-review milestone must assess the actual diff on Standards and Spec axes independently, record concrete findings, and link each mitigation before closure.