Skip to content

Fix acceptance reliability, protocol boundaries and CSV retest lineage - #2

Merged
KaiserIIII merged 2 commits into
mainfrom
codex/quality-audit-20261015
Oct 15, 2026
Merged

KaiserIIII merged 2 commits into
mainfrom
codex/quality-audit-20261015

Conversation

@KaiserIIII

@KaiserIIII KaiserIIII commented Oct 15, 2026 •

Copy link
Copy Markdown
Owner

Acceptance history could be changed by late/conflicting adapter callbacks, and shutdown or storage/startup failures could leave misleading task availability. This AI-assisted review preserves completed evidence, persists retest lineage on the first insert, blocks further operations after terminal persistence failure, and cleans up SQLite bootstrap/worker-start failures.

It also validates HTTP admission/framing before body reads with bounded rejection teardown, applies shared constraints to response alternatives, enforces read-function-specific Modbus limits, and rejects unsupported nested descendants before silently reducing expected-device counts. CSV appends parent_job_id as column 30; the original 29 columns retain their positions and values. Exactly-29-column external consumers need updating. No new dependency or schema migration is required.

Validation:

  • Final exported Git tree: 214 offline tests passed (114 added above the 100-test baseline), 23 original self-checks, configuration validation and compilation passed. Runtime/test bytes match the pushed archive; only changelog/validation prose followed.
  • Independent review reproduced a combined worker-start/storage failure; its regression failed before the minimal fix and passed afterwards. The final storage/lifecycle and HTTP helper follow-up were reviewed.
  • Initial candidate 17b3f4a passed all four push-matrix jobs, but the PR Windows/Python 3.13 job timed out in a preexisting Node helper after 8 seconds (212 other tests passed). Logs are retained. Follow-up 584b5d2 adds request deadlines, exact 200/200/401 plus JSON-failure assertions, a negative server-token control and completion evidence with a separate subprocess budget. The original delay's cause is unproven; it is not claimed as a production bug.
  • Broad runtime line/branch coverage: 50.11%/39.46%. Tests use synthetic adapters, temporary SQLite databases and loopback sockets; no real devices were accessed. Hardware and sustained-load behavior remain unqualified. Final Windows/Linux Python 3.10/3.13 results are in this PR's checks.

AI-assisted reliability review with 113 added offline regressions. Preserve original task evidence, fail closed on persistent storage failure, validate protocol boundaries and append CSV retest parent identity.
@KaiserIIII
KaiserIIII merged commit a291575 into main Oct 15, 2026
8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant