cesnimda
32 lines
3.7 KiB
Markdown
32 lines
3.7 KiB
Markdown
# Review and scope rulings
|
|
|
|
These are implementation/review decisions, not approval to merge or release. Original review
|
|
documents are copied in `reviews/`; they are chronological, so earlier pending results are
|
|
superseded only by the final validation and scoped re-review.
|
|
|
|
| Decision | Rationale / disposition |
|
|
| --- | --- |
|
|
| Requested fix branch in existing checkout | Exact user-requested branch, based on final Phase 9 rather than main; extra checkout and routine approval pauses unnecessary under explicit autonomous authorization. |
|
|
| Shared persistence correction | Reproduced full-snapshot loss requires source revisions plus database predicates. Client-only guards and controller-entry checks cannot protect the eventual commit. |
|
|
| Atomic Career boundary | Injected Identity failure proved partial persistence. Include children/history/projection/import metadata in the same transaction and clear failed tracking. |
|
|
| Related text writers included | Cover-letter, application-drafts and structured tailored-CV SQLite regressions actually lost approved content before repair. Their source guards are directly related scope, not speculative blanket versioning. |
|
|
| Legacy identity CV bypass closed | Projection-only writes could sidestep canonical ownership/revision. Explicit 400 with canonical-route guidance; identity-only fields retain their route. |
|
|
| Narrow lifecycle commands | Job token advances for tracked partial/background mutations, protecting overlapping database changes. Do not invent a full-editor merge contract for field commands or independently owned Settings/company/contact records. |
|
|
| Old callers fail closed | Missing revision cannot safely authorize old snapshots. Update active callers/fixtures explicitly; no blind-save overload, latest-version fetch to bless old bodies, silent merge or automatic retry. |
|
|
| Historical diagnostic files preserved | The old unversioned seed now fails. Retain actual failures and use separately hashed conforming companions with unchanged A/B ordering/volume assertions. No historical assertion is weakened. |
|
|
| Formal RD21 disposition | Original issue requires SQLite AND supported MariaDB runtime proof. Local repair passes, but missing provider qualification means formal UNRESOLVED, not an invented closure exception. |
|
|
| Preservation status | Keep 56 PARTIAL/C11 DEFERRED, all 57 groups/33 routes and manual/provider/output/release obligations. Automated passes do not upgrade their acceptance status. |
|
|
| Distinct remaining work | Creation idempotency and historical recovery unimplemented. Discard-versus-accept extraction lifecycle concurrency considered but unproven; separate evidence needed before a new repair. No new integration or Phase 10 scope. |
|
|
| Tool fallback | WSL Bash unavailable; installed Git Bash ran required skill helpers. Pinned local .NET SDK used rather than the incompatible system SDK. No dependency installation or production workaround. |
|
|
|
|
Early caller review and two backend reviews identified no remaining material issue in their
|
|
bounded scopes. Whole-branch [final review](reviews/final-review.md) found two material integration
|
|
issues: missing CORS response-header exposure and stale Answers entry after sibling writes.
|
|
Both were reproduced with failing assertions, repaired in one wave at `1b89a77` and covered by
|
|
actual browser JavaScript plus native database reads. The
|
|
[scoped re-review](reviews/final-re-review.md) marks both **ADDRESSED**, with no new material
|
|
finding. Reviewers inspected code/evidence; their review is not a second independent execution
|
|
of the coordinator's full suites. Focused 55 Jest/2 browser results overlap final full evidence.
|
|
|
|
The review does not certify every editor, MariaDB, live providers, devices, production or recovery.
|