A pulls list reads every row's status word in one batch: `words` under 5 ms for a page of pulls #6170

The pulls list now words a whole page in one batch, with words under 5 ms for 50 pulls, and each word is the same one pullWord gives today. tsc, the design lint and the pull-words tests pass, as does the new test/pulls-words.test.ts. I did not run the full bun test suite; the forge's check runs it on this push.

What changed

• New pullWords: pullWords(ctx, pulls) in src/web/studio/words.ts replaces the per-pull Promise.all in changesList in src/web/studio/lists.ts. It reuses #6135's knowPulls (checks and reviewers) and adds three reads.
• Added reads: the close reasons of closed pulls are read in one query, so a parked pull still says "parked by the crew". Production's status and the viewer's access to it are read once for the page rather than per merged pull.
• pullWord unchanged: it only prefers these page-level reads when they exist, so every other caller behaves as before.

Tests (test/pulls-words.test.ts)

• Equivalence: 50 pulls cover open with no check, green, failed, running, queued, closed, parked and merged. pullWords is compared with pullWord for each, and every word class appears.
• Timing: the words phase for the same 50 pulls is under 5 ms (best of 5 warm runs).
• Not covered: the test repo is not an app, so a merged pull only reaches "shipped — merged". The production-status path is read once but has no test.

Left out

• runWord on the runs list: it is still per row; this issue only covers pulls.

Closes #6151

The pulls list now words a whole page in one batch, with words under 5 ms for 50 pulls, and each word is the same one pullWord gives today. tsc, the design lint and the pull-words tests pass, as does the new test/pulls-words.test.ts. I did not run the full bun test suite; the forge's check runs it on this push.

What changed

• New pullWords: pullWords(ctx, pulls) in src/web/studio/words.ts replaces the per-pull Promise.all in changesList in src/web/studio/lists.ts. It reuses #6135's knowPulls (checks and reviewers) and adds three reads.
• Added reads: the close reasons of closed pulls are read in one query, so a parked pull still says "parked by the crew". Production's status and the viewer's access to it are read once for the page rather than per merged pull.
• pullWord unchanged: it only prefers these page-level reads when they exist, so every other caller behaves as before.

Tests (test/pulls-words.test.ts)

• Equivalence: 50 pulls cover open with no check, green, failed, running, queued, closed, parked and merged. pullWords is compared with pullWord for each, and every word class appears.
• Timing: the words phase for the same 50 pulls is under 5 ms (best of 5 warm runs).
• Not covered: the test repo is not an app, so a merged pull only reaches "shipped — merged". The production-status path is read once but has no test.

Left out

• runWord on the runs list: it is still per row; this issue only covers pulls.

The full gate on land was red; the pull is open again. Red tests:

• no raw wall-clock budget in a test (#4429) > no test file holds a speed to a raw ms number
• no raw wall-clock budget in a test (#4429) > no test file holds a speed to a raw ms number [390.89ms] [shard 703]

The pulls list now reads every row's status word in one batch (pullWords), and the speed check in test/pulls-words.test.ts no longer uses a raw ms number. That raw 5 ms assertion was what failed the "no raw wall-clock budget in a test" (#4429) gate. test/pulls-words.test.ts, test/flake.test.ts and test/budget.test.ts pass and tsc --noEmit is clean, but I haven't run the #4429 gate itself.

• Change: the 5 ms bound now goes through expectBudget, judged against a baseline measured in the same run (the best of 20 trivial DB reads). The absolute 5 ms holds only when FORGE_TEST_QUIET=1. Otherwise the limit scales with the baseline, capped at 4× the budget. The test that checks pullWords returns the same word as pullWord for all 50 pulls is unchanged.
• Not done: @juniper asked for a test of the production-status path, including the null case for a guest on a private app. I didn't add it. It needs a repo that is an app, which this test file doesn't build, and I'd rather say so than half-cover it. That path is still untested, so it's the follow-up to pick up.

I added the missing test for the private-app gate in pullWords, and the branch is pushed. bun test test/private-app.test.ts passes (14 tests). I did not re-run tsc --noEmit after the change.

• What it covers: the test is in test/private-app.test.ts, on the private app that file already sets up. For a page of two merged pulls, pullWords must equal pullWord row for row. That is checked for the owner, the crew bot, a guest (carol) and a signed-out viewer, and for production serving the merged commit versus production failing on a different one.
• The gate: the guest and signed-out viewer must read { word: "shipped", note: "merged" }, and SECRET-ERROR must not appear in what they read. The owner reads "production is serving it" for the caught-up pull and something other than "merged" for the one that isn't.
• It fails without the gate: I removed the canEnterApp check at src/web/studio/words.ts:189. The new test failed and the other 13 passed. I then restored the file.
• Not covered: a public app's production path is not tested.

Steward tick: this pull is red on its own merge with current main (gated alone, failed) and was held out of the landing queue so the others can land. Rebase on main, fix the failing test, push - then the hold comes off.

Steward tick: this pull is red on its own merge with current main (gated alone, failed) and was held out of the landing queue so the others can land. Rebase on main, fix the failing test, push - then the hold comes off.

The full local run was red in many places that have nothing to do with this pull, so it tells me nothing about the gate. I can't run the sandbox-skipped suites here, and the machine is nearly out of disk. Those failures are things like Chromium captures, the SSH port, the merge queue, capacity and report mode.

I haven't changed anything since the last push. The branch has main merged in (0b3b2a27), and tsc --noEmit and the tests that cover pullWords pass. Those are pulls-words, private-app, flake and budget.

If the forge's gate on this push is still red, send me its log and I'll fix the failing test.

A pulls list now words every row from one batched read of checks, close reasons and production's status. words for 50 pulls stays under 5 ms. This push rebases the pull on main, speeds up its test, and adds the missing public-app test.

bun test test/pulls-words.test.ts test/private-app.test.ts and tsc --noEmit passed before I merged main and made the last test edits. After those edits only test/pulls-words.test.ts was re-run, and it passes. I did not re-run the private-app test or the type check on the merged tree.

Changes since review

• Rebase: I merged current main into the branch. The pull was red alone against main, so this puts the gate on its own merge.
• Test cost: test/pulls-words.test.ts no longer makes branches, commits or merges through git, and the 180 s beforeAll timeout is gone. It inserts 50 pull rows directly, one of each state: waiting, look, stopped, working, shipped, parked, closed. The file runs in about 0.6 s.
• Public-app path: a new test gives a public app a fake production status and reads it as an anonymous viewer. It checks three things:

- Production is read once for the whole page. - A pull production has caught up to says "production is serving it". - Later merged pulls say "merged; waiting for production to redeploy". - Each batched word equals the word pullWord gives for that row.

Direction: I disagree with the "parked" point only in part. #6151 is a speed change to an existing list and adds no feature. But if a person wants it held while the repository is flow-only, close it and reopen it when the direction lifts. I did not park it myself.

Left out: no markup changes, so no design lint run.