cesnimda
3.4 KiB
Phase 9 defect register
Status is evidence-specific. Historic acceptance records remain unchanged. A passing characterization of a known limitation does not make the limitation safe.
| ID | Severity / impact | Reproduction and disposition | Owner / closure | Phase 9 / production |
|---|---|---|---|---|
| P9-T01 | Medium test reliability: parser descendant test could time out before it had a descendant | Entry sidecar 55 pass/1 fail (missing PID); isolated run passes; explicit slow startup reproduces the race. Fixture now waits for its child PID before the unchanged 200 ms deadline. Real process-tree kill and HTTP 504 assertions retained. Focused green and full 56/56 successor. No parser product change. | QA; .artifacts/phase9/sidecar-entry-failed.*, sidecar-fixture-red.*, sidecar-fixture-green.*, sidecar.xml |
Resolved in tested Windows environment; Linux run remains separate. |
| P9-L01 | Low localisation defect: Admin grid vendor controls appear in English in Bokmål | Sole DataGrid owner omitted localeText. New English desktop test passes while Bokmål fails to find “Rader per side:”. Reuse installed nbNO locale in AdminUsersPage. Focused 12/12 green; complete regression/native successor required. |
Frontend; admin-users-page.test.tsx, AdminUsersPage.tsx; retained RED/GREEN logs/JSON |
Pending integrated successor; no authorisation, pagination or data contract change. |
| RD21 | High data-integrity limitation: generic job/canonical Career stale saves and creation-intent idempotency lack the complete CV-style revision contract | Inherited, not automatically closed by preference/CV conflict tests. Source/previous runtime evidence and any fresh characterization must be read separately. Do not infer server expected-version protection from a component dirty-state guard. | Domain owners; separately scoped contract/design, all callers, server concurrency and recovery tests. No silent architectural expansion in QA. | Retained qualification; blocks unqualified concurrent-write/production claims. |
| P9-PERF-01 | Low duplicate report work: shell and Dashboard each initially read reminders | Phase 8 source/network diagnosis retained. Phase 9 request-count evidence must distinguish initial duplicate reads from additional reads during editing. | Shared reminder read owner; measured benefit and independent shell/dashboard failure/loading tests before consolidation. | Accepted existing development debt; no Phase 10 implementation here. |
Failed diagnostic fixtures are also retained: initial route matrix mistook “CV unavailable” alert text for a heading; correct page heading is “Public CV”. Initial sustained fixture compared submitted JSON directly with the canonical normalizer and used derived date fields; corrected fixture uses canonical start/end, distinct bullets and a verified saved readback before checking preservation. The initial drawer assertion inspected immediately after ARIA removal rather than waiting for the closing transition; settled CSS/keyboard proof is required. These are QA harness mistakes, not demonstrated product defects, and failed attempts are never counted as passes.
The ordinary run interrupted after 36 completed cases is not an accepted complete run. Native-command warnings were initially promoted to terminating PowerShell errors; successor wrappers preserve actual native exit codes. No assertion was suppressed, no configured retry added and no skip introduced.