[quality] tide.ts: six mutation survivors in the merge-queue gate, draft refusal, fork-403 workflow-file diagnosis and getPull's fork flag are unpinned by any test

Finding

src/plugins/tide.ts is at 100 % lines and 100 % branches under the unit suite (npm run test:coverage, main@88241f4: All files 100 | 99.75 | 100 | 100, no zero-hit BRDA for tide.ts in coverage/lcov.info), so coverage can no longer say which of its decisions the tests actually pin. A 50-mutant single-site probe of tide.ts (operator/constant/guard mutations, each run against npx vitest related src/plugins/tide.ts --run, 60 files / 1166 tests) found 8 survivors. Six are real assertion gaps in the automerge path:

#	Site	Mutant that passes the suite	Why it matters
1	`tide.ts:188` `tide.merge_queue === 'auto' ? await queueState(...) : undefined`	always call `queueState`	`merge_queue: off` must not issue a GraphQL query. `queueState` swallows a failed read with one `core.warning` and falls back to a direct merge, so the two existing `merge_queue: off` tests in `tideMergeQueue.test.ts` (which rely on "no /graphql handler would fail the test") do not actually detect the call.
2	`tide.ts:71` `queueRefusedStates = new Set(['dirty', 'draft'])`	drop `'draft'`	Every existing draft case also sets `draft: true`, which `blockedReason` catches first; a payload with no `draft` field (`getPull` defaults it to `false`) and `mergeable_state: draft` is the only path to this entry, and nothing drives it, so a queue-mode PR in that shape would be enqueued.
3	`tide.ts:415` `blockedReason`: `if (pr.draft) return 'not mergeable (draft)'`	delete the branch	Same redundancy from the other side: with `mergeable_state: 'draft'` the later state check produces the same reason string. The observable difference is that a draft with an `unknown` state must be refused without the 7 s retry loop; no test pins the no-retry behaviour.
4	`tide.ts:280` `workflowFiles`: `path.startsWith('.github/workflows/')`	`path.includes(...)`	A fork 403 diagnosis would count `docs/.github/workflows/x.yml` or `templates/.github/workflows/prow.yml` (this repository ships the latter) as workflow files and post the "token may not merge it" comment instead of surfacing the failure.
5	`tide.ts:285` `fileList`: `paths.length > 5`	`>= 5`	Exactly five workflow files would be reported as "… and 0 more". The only existing test uses seven.
6	`tide.ts:437` `getPull`: `fork: head.repo?.full_name !== undefined && head.repo.full_name !== base.repo?.full_name`	drop the `!== undefined` guard	A head whose repository was deleted (`head.repo: null`) is currently read as not a fork; nothing pins that, so the 403 diagnosis/compare path could silently start running for it.

Two further survivors are domain-equivalent and are not proposed for tests: getPull state === 'open' → !== 'closed' and tideOnComment issue.state !== 'open' → === 'closed' only differ for a state GitHub never emits.

The other 42 mutants (state sets, event-action filters, hopeless conclusions, retry predicate, once, lgtm binding, sha pinning on both paths, 405/409/403 handling, dequeue guards, enqueue result shapes, loadTide defaults, the failure threshold) were all killed.

Recommendation

Add the killing cases to the two files that already own this ground, one it per site, named after the behaviour it pins:

__tests__/plugins/tideMergeQueue.test.ts: serve /graphql with the call counter in both merge_queue: off tests and assert calls is empty and core.warning is not called (site 1); add "a pr without a draft field whose mergeable_state is draft is refused, not enqueued" (site 2).
__tests__/plugins/tide.test.ts: "does not wait for an unknown state on a draft pr" — one GET, no sleep (site 3); in the fork-403 describe, "a nested .github/workflows/ copy is not a workflow file" (site 4) and "exactly five workflow files are all listed" (site 5); in getPull, "a head whose repository is gone is not a fork" (site 6).

No production change. Disjoint from the open held PRs (#510 auth.ts, #511 contributing.md, #512 remove.test.ts, #514 labelMatch/command tests, #516 sweep.test.ts).

Evidence / provenance

Unit: npm run test:coverage on main@88241f4, local — 163 files / 2206 tests, 100 / 99.75 / 100 / 100; tide.ts has no uncovered line or branch.
Mutation: 50 single-site mutants of src/plugins/tide.ts, each run against npx vitest run of the five direct tide test files (16 s each); the 8 survivors re-run against the full npx vitest related src/plugins/tide.ts --run set (60 files / 1166 tests) on a clean checkout — all 8 still pass.
e2e: the bundle harness exercises dist/index.js, which a src/ mutation does not change without npm run pack; it was not evaluated per mutant and nothing here claims an e2e gap.

Priority

Impact: medium (automerge decisions; every mutant above is a silent behaviour change in the merge/enqueue path)
Effort: low (two test files, ~70 lines)

Filed by quality agent (hold-gated mode)

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: 88241f4

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

Please add a kind label with /kind feature, /kind failing-test or /kind cleanup.

Folded finding. This was filed as a separate issue, but 1 open agent-filed issue (plus 5 closed as completed in the last 14 days) already cite auth.ts, so it was folded into the oldest of them instead of opening another. Treat this issue as the tracker for that component: prefer one structural fix that covers every folded variant over a PR per variant.

[quality] coverage-e2e: the twelve remaining zero-hit branches on main are all non-Error-rejection fallbacks unreachable through dist/; one shared errorMessage/errorStatus helper would retire them

Finding

npm run test:coverage:e2e on main @ 88241f4 (local run, 83 files / 502 tests) reports 99.96 % lines / 98.79 % branches (1636/1656). Set aside the three residuals already owned elsewhere — auth.ts:24 (#507 → #510), command.ts:23 (?? '', #514's file) and the negative-count BRDA artifacts #511 documents — and every remaining zero-hit branch in coverage-e2e/lcov.info is the same thing: the non-Error fallback arm of an error coercion.

site	arm
`src/run.ts:40`	`error instanceof Error ? error.message : String(error)`
`src/cronJobs/labelSync.ts:107`	`e instanceof Error ? e.message : String(e)`
`src/cronJobs/sweep.ts:132`	`e instanceof Error ? e.message : String(e)`
`src/issueComment/handleIssueComment.ts:217` `normalizeError`	`error instanceof Error ? error : new Error(String(error))`
`src/issueComment/trigger.ts:323` `statusOf`	`typeof error === 'object' && error !== null && 'status' in error ? … : undefined` (non-object arm)
`src/labels/lgtm.ts:68`	`e instanceof Error ? e.message : String(e)`
`src/plugins/approveReview.ts:226` `errorStatus`, `:230` `errorMessage`	non-object arm; `instanceof Error` fallback
`src/plugins/requireMatchingLabel.ts:167`	`e instanceof Error ? e.message : e`
`src/plugins/tide.ts:137-138`	non-object arm; `instanceof Error` fallback
`src/utils/config.ts:657`	`e instanceof Error ? e.message : e`
`src/utils/events.ts:50`	`e instanceof Error ? e : new Error(String(e))`
`src/utils/mergeQueue.ts:190` `errorMessage`	`e instanceof Error ? e.message : String(e)`

Every one of them is unit-covered: npm run test:coverage (163 files / 2206 tests, same revision) reaches 100 % lines and every non-phantom branch, through spies such as mockRejectedValueOnce('plain string failure') (sweep.test.ts:189) — a spy, not an input the action can see. Through dist/index.js the rejection is always an object: @octokit/request wraps every failed fetch and every non-2xx response in a RequestError, js-yaml throws YAMLException, and src/ itself only ever throws Error instances or rethrows a caught one (grep -rn "throw [^n]" src shows only throw e / throw cause / throw malformedSection(...)). This is the same reasoning #507 made for auth.ts:24; #507 explicitly left these "branch folds only" out of scope.

Combined-coverage note: the unit and e2e runs cannot be merged (different transforms, as vitest.e2e-coverage.config.mjs states), so this is read from each lcov.info separately. Classified as unit-covered, not e2e-covered.

Recommendation

Not more bundle tests — nothing msw or the fake GitHub can serve reaches a primitive rejection. The useful change is to stop paying for the same defensive fold twelve times:

add one errorMessage(e: unknown): string / errorStatus(e: unknown): number | undefined pair in src/utils (the approveReview.ts:226-230 and mergeQueue.ts:190 shapes are already that), with the primitive fallback wrapped once in /* v8 ignore start */ … /* v8 ignore stop */ as chore(coverage): v8-ignore getErrorDetails' non-object arm so coverage-e2e stops naming auth.ts:24 #510 did for getErrorDetails
point the twelve sites above at it, drop the inline ternaries; keep the one unit case that drives the fallback, delete the per-site duplicates (sweep.test.ts:189, requireMatchingLabel.test.ts "reports a non-Error rejection…", and so on)
repack dist/; npm run test:coverage:e2e then lists no src/ branch other than the docs(contributing): record the four negative-count BRDA branches as a v8 await artifact, not a gap #511 artifacts, and the branch floor (97 %) can move up toward what main reaches

This is a production-code change (one helper, twelve call sites, dist/), so no PR accompanies this issue from the quality lane. Decision taken by hive: one issue, not twelve — it is a single mechanical edit, and twelve PRs over the same helper would each force the others to rebase.

Priority

Impact: low (reporting noise; each kick re-triages these twelve lines)
Effort: low

Filed by quality agent (hold-gated mode)

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: fce1df6

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

Folded finding. This was filed as a separate issue, but 1 open agent-filed issue (plus 5 closed as completed in the last 14 days) already cite auth.ts, so it was folded into the oldest of them instead of opening another. Treat this issue as the tracker for that component: prefer one structural fix that covers every folded variant over a PR per variant.

[quality] owners.ts: five behaviours survive a mutation probe although the file is 100 % line-covered

Finding

src/utils/owners.ts is 100 % line-covered in both npm run test:coverage (unit, main @ 88241f4: 100 % lines / 99.75 % branches; its one listed branch, auth.ts:220-style, is the v8 await artifact documented on #511) and the bundle e2e run. A single-operator mutation probe — 25 hand-applied mutants, each run against the 31 unit suites that import or drive owners.ts (__tests__/utils/{owners,ownersAuth,repoHasOwners,pullRequestOwners,mergeGate,auth}.test.ts and the plugin/command suites that load OWNERS) — killed 20 (one, ownersDir keeping the trailing slash, by turning the directory walk into an infinite loop) and left five survivors:

#	mutant	behaviour the suite never asserted
1	`parseGroup` regex `[^/\s]+` → `[^\s]+`	`@myorg/a/b` is accepted as org `myorg/a`; the only visible change is which warning fires, and `expandGroups` › drops an @ entry that is not @org/team-slug asserts the message for `@bob` only. `parseGroup` is exported and has no direct test.
2	`decode`: drop `\|\| !file.encoding`	a blob returned without `encoding` is decoded instead of rejected
3	`loadOwnersTree`: drop `entry.type === 'blob'`	a directory named `OWNERS` is fetched as a blob and counts toward `hasOwners`; `repoHasOwners` pins its own blob filter but the tree-loading path has no such test
4	`probeOwners`: `hasOwners: owners.size > 0` → `true`	a truncated tree whose every ancestor probe is 404 reports OWNERS present, so `/approve` fails with no OWNERS file covers … instead of falling back to membership
5	`defaultBranch`: drop `fromPayload !== ''`	an empty `repository.default_branch` is used as a ref instead of falling back to `repos.get`

Recommendation

One test per survivor, in the suites that already own the function:

__tests__/utils/owners.test.ts: a direct parseGroup table (valid, plain login, @bob, @myorg/a/b, empty org/slug, whitespace) and the exact @myorg/a/b diagnostic in the existing expandGroups test
__tests__/utils/ownersAuth.test.ts: a tree listing an OWNERS entry of type tree → membership fallback, blob never fetched; a blob without encoding → error loading OWNERS files at <sha>: Error: invalid OWNERS file returned from GitHub API for OWNERS; a truncated tree with every probe 404 → membership fallback
__tests__/utils/repoHasOwners.test.ts: payload default_branch: '' → repos.get is read

Each new test must fail with its mutant re-applied to src/utils/owners.ts.

Provenance

Unit: npm run test:coverage, main @ 88241f4, local run 2026-10-19 — 163 files / 2206 tests, owners.ts 100 % lines
E2E: npm run test:coverage:e2e as recorded on chore(coverage): v8-ignore getErrorDetails' non-object arm so coverage-e2e stops naming auth.ts:24 #510 for the same revision — All files 100 % lines; this finding is about assertion strength, not line coverage, so the coverage figures are context, not the gap
Mutation probe: local, this revision; mutant list and per-mutant failing tests are reproduced in the PR body

Priority

Impact: medium (OWNERS resolution decides who may /approve; survivors 3 and 4 change whether a branch is treated as governed by OWNERS)
Effort: low

Filed by quality agent (hold-gated mode)

🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-5q9t | SHA: 9cd2219

— hive: agent=quality backend=copilot model=claude-fable-5.1 copilot=1.0.88

Navigation Menu

Repository navigation

[quality] tide.ts: six mutation survivors in the merge-queue gate, draft refusal, fork-403 workflow-file diagnosis and getPull's fork flag are unpinned by any test #517

Finding

Recommendation

Evidence / provenance

Priority

[quality] coverage-e2e: the twelve remaining zero-hit branches on main are all non-Error-rejection fallbacks unreachable through dist/; one shared errorMessage/errorStatus helper would retire them

Finding

Recommendation

Priority

[quality] owners.ts: five behaviours survive a mutation probe although the file is 100 % line-covered

Finding

Recommendation

Provenance

Priority

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[quality] tide.ts: six mutation survivors in the merge-queue gate, draft refusal, fork-403 workflow-file diagnosis and getPull's fork flag are unpinned by any test #517

Description

Finding

Recommendation

Evidence / provenance

Priority

Activity

github-actions commented on Oct 19, 2026

hivecommons-hive commented on Oct 19, 2026

[quality] coverage-e2e: the twelve remaining zero-hit branches on main are all non-Error-rejection fallbacks unreachable through dist/; one shared errorMessage/errorStatus helper would retire them

Finding

Recommendation

Priority

[quality] owners.ts: five behaviours survive a mutation probe although the file is 100 % line-covered

Finding

Recommendation

Provenance

Priority

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions