test(renderer): give the repeated project turn link test an explicit 15s budget

[PeterGuy326]

Preflight

• Searched open and closed PRs for related changes (ci(renderer): the repeated project turn link test times out on fresh runs of main (macos-14) #532 tracks this; ci(renderer): App context navigation toolbar test times out on fresh runs of main (both platforms) #529/test(renderer): give the full-App toolbar test an explicit 15s budget #531 are the identical fix for the toolbar test in this same file; fix(ci): Node 24 / ubuntu-latest fails on the service-connections PAT token assertion #512/fix(services): keep unsaved connection drafts across background refreshes #513 cover the separate macOS PAT assertion).
• No credentials, personal data, private paths, or private screenshots.
• Test-only change on top of main @ ed94fca.

Summary

reanchors a repeated project turn link without replacing the conversation draft renders the full App, opens the project workspace and re-anchors the same turn twice inside a for loop, so its cold cost sits above the 5s vitest default on loaded runners. Fresh runs of main currently fail on Node 24 / macos-14 with Test timed out in 5000ms (#532), which keeps that required check red for every open PR.

This gives the test an explicit 15s budget with a comment recording why — the same treatment #531 gave the toolbar test, and matching the other full-App tests in this file that already carry 10_000/15_000 budgets.

Evidence

Where	SHA	Run / job	Result
main	ed94fca	run 36518952016, job 109247408016 (Node 24 / macos-14)	Test Files 1 failed | 70 passed (71), Tests 1 failed | 884 passed (885), App.test.tsx:2667:1 → Error: Test timed out in 5000ms.
#527 branch	b6cedd3	run 36517817744 attempt 1, job 109243938638 (Node 24 / macos-14)	identical signature
#527 branch, same run	b6cedd3	run 36517817744 attempt 1, Node 24 / ubuntu-latest	success — same commit, same code
#527 branch, plain rerun	b6cedd3	run 36517817744 attempt 2	success on both platforms

Same commit passes on ubuntu in the same run, and passes on macOS on a rerun with no code change — so the test sits at the 5s edge and loses under runner load. It is not a hang and it is not caused by any PR's diff: the failing test never touches the thought-trail code path.

Assumptions

• The test is slow, not hung. A hang would have failed the rerun and would have failed ubuntu-latest in the same run.
• Raising the per-test budget is the right fix rather than lowering the gate: this is a test-harness budget, not a product timeout. No production code path changes.
• 15s is the right number for consistency — the other full-App tests in this file already use explicit 10_000/15_000 budgets, and this is exactly what test(renderer): give the full-App toolbar test an explicit 15s budget #531 set for the toolbar test one function above.

Verification

• Local, this branch: vitest run --config <temp> renderer/test/App.test.tsx -t "reanchors a repeated project turn link without replacing the conversation draft" → Tests 1 passed | 110 skipped (111), the test itself in 1441ms.
• So it is comfortably fast on an idle machine and only blows the default under CI load, which is what the four rows above show.

Acceptance

• Node 24 / macos-14 and Node 24 / ubuntu-latest both green on fresh runs of this branch and of main after merge.
• No other test in App.test.tsx changes behaviour: the diff is one added comment plus }); → }, 15_000); on a single it.

Found during the 2026-09-29 duty sweep while checking why #527's macOS job failed while ubuntu passed in the same run.

Closes #532.

[PeterGuy326]

Third independent occurrence, this time on #527 after its branch update — same test, same signature, and ubuntu-latest green in the same run again:

Where	SHA	macOS job	Result
main	ed94fca	109247408016	App.test.tsx:2667:1 → Test timed out in 5000ms
feat/qoder-thought-trail (attempt 1)	b6cedd3	109243938638	identical
feat/qoder-thought-trail (after update-branch onto ed94fca)	fea69e7f	109251809950	identical

Three different merge bases, three different runs, the same single test in App.test.tsx, with Node 24 / ubuntu-latest green in every one of them. That is the whole argument for this PR: per-test budget on a loaded macOS runner, not something a branch changed. The test passes locally in 1441ms.

I have rerun the fea69e7f macOS job so #527 is not stuck behind it in the meantime — but the flake will keep coming back until this budget lands.

[PeterGuy326]

Measured, not inferred — the test is 7.8% over budget, and the file routinely runs tests at 4–8s

I re-ran the fea69e7f macOS job and it failed again on the same test, which gave me a clean timing sample from a failing run (109254736100). Vitest prints the elapsed time even for the timed-out test, so the number is right there:

Test (same file, same run)	Elapsed on macos-14	Budget
reanchors a repeated project turn link without replacing the conversation draft	5389ms	5000ms default → failed
App context navigation > consolidates the collaboration toolbar and places employee actions beside the tabs	4199ms	15000ms (added by #531) → passed
restores the original workspace's running task and cancels its exact owner after navigation	8436ms	explicit budget → passed
preserves the actual composer, draft and selected session across organization graph navigation	6135ms	explicit budget → passed
runs A/B/C independently and keeps late responses, streams and cancellation in their own sessions	5523ms	explicit budget → passed
drops workspace A's late group 202 after switching to B and reloads A on return	4561ms	explicit budget → passed

Whole file: App.test.tsx (111 tests | 1 failed) 160081ms on macOS.

What this settles

• Not a hang. The test consumed 5389ms of real work and only overshot by 389ms. A hang would report ~5000ms while doing nothing and would fail on ubuntu-latest too — ubuntu-latest is green in this same run.
• Not this branch's code. The test never touches the thought-trail path.
• The 5s default is simply too tight for this file on macOS runners. Five sibling tests in the same run needed 4.2–8.4s and only pass because they carry explicit budgets. This one was the last heavily-loaded full-App test still sitting on the default.
• 15s is ~2.8x the observed cost, matching what test(renderer): give the full-App toolbar test an explicit 15s budget #531 chose for the toolbar test directly above it and the other full-App tests in the file.

Since opening this PR the test has now failed 4 times in a row on macos-14 (109247408016 on main@ed94fca, 109243938638 and 109251809950 and 109254736100 on the #527 branch) and passed on ubuntu-latest every single time. It is currently the only thing keeping main and #527 red.

[PeterGuy326]

Superseded — closing this rather than rebasing.

While this PR was open, #534 (97656cdc, test(renderer): explicit budgets for the remaining full-App edge tests) landed the same change on main, covering this exact test:

  // Full-App render with two module switches and turn-link re-anchors, so on
  // slower runners it sits above the 5s default (#529). 15s matches the
  // other full-App tests in this file.
  it("reanchors a repeated project turn link without replacing the conversation draft", async () => {
...
  }, 15_000);

#534 went further and also fixed the HireDrawer binding and the docs undo/redo test from the same family, so there is nothing left here that main does not already have — and the branch now conflicts on the same lines (mergeable_state: dirty). Rebasing would just produce two identical diffs racing for the same lines.

The timing evidence in the two comments above still stands as the measurement behind the fix, and applies equally to #534's change: 5389ms actual against a 5000ms default on macos-14, with five sibling tests in the same run needing 4.2–8.4s, and ubuntu-latest green in every run where macOS failed.

Branch fix/532-project-turn-link-test-timeout deleted.