[polewskm]

Calibrates test-gap severity in the specialist prompt, the tests charter and synthesis: a missing test is should-fix only when the untested path is new behavior with a plausible user-visible failure and no test below it covers it; a wrapper over tested pieces, a test that only pins current behavior, or a second test for a covered path is a nit.
Eval concerns can now name a minSeverity (judged only against findings at that severity or above), and five cases from #30 cover both sides: two expected should-fixes (the inline newest-summary choice at 1c8c1bd, the unasserted POST body at 9f052ed) and the three threads cited in the issue as must-not-be-should-fix.
The paid eval (npm run eval) was not run; npm run typecheck and npm test pass.

Decisions:

• Must-not concerns use minSeverity: should-fix, so the bot may still raise these gaps as nits; the eval would otherwise forbid them outright.
• The two should-fix examples are Post a new summary comment for each review #30 test findings the maintainer fixed where the logic had no test below it, as the counterpart to the three inflated ones.
• Each case reviews the PR base against the head the original finding was raised on, with the PR body as it read at that time.

Closes #37

[ncode-review]

🤖 NCode Review at 1840ef6: 5 new findings on the changed lines.

[ncode-review]

🤖 Review complete

Verdict: ✅ Approve — No findings.

5 reviewers run — 0 findings (run details)

Reviewer	Model	Findings	Top
conventions	sonnet-5-5	0	🟢
correctness	sonnet-5-5	0	🟢
hygiene	haiku-5-5	0	🟢
intent	haiku-5-5	0	🟢
rule-check	sonnet-5-5	0	🟢

Skipped: api-compat, architecture, completeness, data-model, infrastructure, performance, security, tests

_{Trial: findings are posted as review threads and in the check run. $0.31 on Amazon Bedrock, 30s, at 502f569. (how to interact)}

[polewskm]

/review

[ncode-review]

🤖 NCode Review at 7aa264e: 1 new finding on the changed lines.
2 more were already posted on this pull request.

[polewskm]

Autopilot: answering the review summary at #65 (comment)

• F1 (reviewers/tests.md, the rule worded differently in several prompts): declined. This repeats an earlier finding. See the thread reply at Rate test gaps over tested wrappers and covered paths as nits #65 (comment) and the first answer at Rate test gaps over tested wrappers and covered paths as nits #65 (comment).
• F2 (CHANGELOG.md, the PR title isn't a conventional commit and there's no Closes #37): declined. The PR description already ends with Closes #37. This repository doesn't use conventional-commit titles: the merged PRs (Follow up the engine fetch's review: a fetch time limit, a testable skip decision, sturdier tests #22–Enable-CodeReview: -SkipBedrock still grants the Bedrock key, and -Repo splits on commas #60) are titled in plain imperative sentences, as this one is.
• F3 (evals/run.ts, candidate and Concern exported, and a main-module guard added, only for tests): declined. candidate is the minSeverity filter this PR adds, and exporting it with the standard main-module guard is the smallest way to test it. Moving it into its own module would widen the PR for a nit.

[polewskm]

Autopilot: Round 2 answered. No blocker or should-fix is left open; the remaining threads are nits, answered above. Recommend merging once you've resolved them.