V4.3 Regression Coverage
V4.3 regression coverage
Issue evidence includes the accepted Issue #197 comment.
Consolidated scaffold baseline delivery research: d2ce164aee1f59755fce84cbff40afed9e4ed087.
Accepted scaffold coverage is AC-037–AC-045: IP-463 owns the complete capability, archive, reader, and ownership slice; IP-464 owns health; IP-466 owns proof recovery; IP-468 owns scoped invalidation; IP-470 remains the Issue #197 candidate fix. The runtime baseline.dataRoot/shared-agents.md and archive data/shared-agents.md are two path representations of the same selected asset, and regression evidence must distinguish them.
This record makes the accepted shared iteration traceable to GitHub Issue #197 and the distributed consumer .agents/AGENTS.md refresh concern. It records coverage already accepted by the scaffold lifecycle specification; it adds no product criteria, experiments, or task graph edges.
| Existing work | Regression coverage |
|---|---|
IP-470 | Covers Issue #197's candidate path: AC-033–AC-036 prove that a workspace-local effect-tsgo available under the ordinary planned installation resolves in isolation, preserves containment and disabled scripts, and permits update under satisfied preconditions. The underlying consumer failure cause remains un-reproduced. |
IP-463 | Covers baseline-owned shared .agents/AGENTS.md refresh: OUT-001/003/006 and AC-001, AC-008–AC-011 detect stale or missing resolved authoritative shared managed guidance and refresh it through an authorized successful update while preserving authored root/scoped guidance and custom roles. Ownership is evaluated by the declared contract; every agent prompt is not treated as one byte-equality target. |
IP-464 | Covers AC-006/AC-007: check reports independently assessable stale shared managed guidance even when an unrelated quality or hook component is blocked, and never reports current without proof. |
IP-466 | Covers AC-017–AC-020: a failed operation cannot claim refreshed shared-file bytes or adopted proof for an unapplied file; completed independent actions and truthful pending evidence remain distinguishable. |
Positive refresh, subsequent no-op, protected authored-byte, and failed-candidate truthful-proof cases are regression obligations derived from accepted scaffold testing decisions; proof remains pending. The reported consumer repositories are collective-intelligence and possibly ci-emera; readonly lookup found /Users/stefan/Desktop/repos/emera, while ci-emera was absent. Exact invocation and outcome remain unknown. Differing current and source hashes alone do not establish the bug because source delivery and older generation may differ.
Candidate implementation seams are common planning ownership, where explicit ownership is classified once and reused for check, diff, and write, and candidate topology repair at the authoritative planned-installation boundary. Keep baseline source availability, candidate refusal, and shared-prompt preservation as separate possible causes until evidence joins them.
For IP-463, the selected baseline is now the accepted content source for the complete scaffold-owned .agents/AGENTS.md. This supersedes the earlier proposal-only status. The installed CLI lookup is current code behavior that implementation must replace. Current code evidence shows update/run.ts classifies existing agent-prompt files as project-authored before comparison and apply filtering, while prompts.ts loads the shared template from the installed CLI data root (with data.ts defining that root). scaffold/output.ts remains part of the exact published source set. These facts motivate the regression trace; they do not establish the consumer bug's cause.
Existing eighteen-task graph and twenty blockers remain unchanged. The historical V4.4 receipt remains historical; current iteration placement follows ITERATION-COORDINATION.md.
Source: accepted scaffold specification d2ce164aee1f59755fce84cbff40afed9e4ed087 and accepted iteration coordination record ee6caf90b364a4720890a0d5063b3972291463fe.