[objectstack-fleet]

Requested by Elon · project thread

Part of #25924 (item 2 of the card's Recommendation: the coverage lane on a schedule, thresholds unchanged). Part of #25677 (roster row 7 of #25697, "Test (coverage shard n/8) · Test (coverage)": kept, follow-up port).

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么: 新增根 workflow ui-coverage-nightly.yml(名为 UI Coverage),把 objectui 的覆盖率车道(8 个分片 + 1 个合并任务)搬回根目录,只在每晚 06:13 UTC 定时跑,也可以手动触发。阈值不变(lines 40、functions 33、branches 30、statements 40),仍由合并任务在完整报告上检查。
• 为什么: 卡上说的「每次 push 都跑」在合并后的仓库里已经不成立:ui/ 合并时这条车道没有移植,ui/.github/ 里那份 GitHub 不会执行,所以从合并那天起这四个阈值没有任何地方在检查。现在按卡的建议以定时方式补回,不回到每次 push。
• 谁会发现红: scheduled-full-run-card.yml 现在也监听 UI Coverage,定时运行红了就开或刷新一张卡 nightly full run: red on main (UI Coverage)。
• 不变的: 不进必需检查,不进发版门(RELEASE_GATE_WORKFLOWS 仍只有 CI 和 Lint & Type Check),PR 和合并队列都不跑它。
• 代价: 覆盖率掉到阈值以下要到第二天早上才报出来(卡上已接受);每晚约 8×22 分钟 runner。
• 卡上的另一项(install 重试)没有在本 PR 里做,原因见下文,交席位定。
• 受管面: .github/workflows/**,Tier H,需要 GOVERNED_APPROVERS 批准。

What changed

File	Change
.github/workflows/ui-coverage-nightly.yml	New. ui-coverage (Test UI (coverage shard n/8), 8 shards, 40 min): install, the seven-package build ci.yml's ui jobs use, then pnpm --dir ui run test:coverage with --reporter=blob --reporter=default --reporter=github-actions --shard=n/8 --testTimeout=60000 and the thresholds zeroed on the legs; uploads ui/.vitest-reports/*. ui-coverage-report (Test UI (coverage)): refuses a partial blob set, merges with --merge-reports (where the configured thresholds are enforced), publishes ui/coverage/, prints the shard headroom (report-only) and states whether the gate ran. Triggers: schedule 13 6 * * * and workflow_dispatch. Caches are restore-only (Build UI Deps' turbo seed, the pnpm store).
.github/workflows/scheduled-full-run-card.yml	Watches UI Coverage too.
scripts/ci/scheduled-full-run.mjs	CARD_CADENCES['UI Coverage'] = 'nightly' (the nightly wording already existed). Self-test +4 cases: the nightly identity, the nightly body, the workflow's single daily cron, and the watcher list and the cadence map naming the same workflows. Battery floor 42 → 46.
ui/scripts/__tests__/coverage-shard-width.test.ts, coverage-shard-reporter-readability.test.ts	Restored from before #25879 (they retired with the lane), now reading the new file through helpers/root-ci.ts.
ui/scripts/__tests__/shard-headroom.test.ts	The coverage half of the retired wiring pin comes back against ui-coverage-report; the test-aggregate half stays retired (no root job).
ui/scripts/__tests__/helpers/root-ci.ts, ui/scripts/shard-headroom.mjs	UI_COVERAGE (file + job keys); header prose.

Measurements

• The coverage lane's trigger on the merged repo: no root workflow runs test:coverage for ui/ (grep -rn 'test:coverage' .github/workflows/ → only coverage-nightly.yml, which is spec's). objectui's test-coverage / coverage-report jobs exist only in ui/.github/workflows/ci.yml, which GitHub does not run. The ui coverage thresholds have had no reader since the cutover.
• Readers checked before adding the lane: required contexts (branch ruleset and classic protection: none of the new names, and none is added); RELEASE_GATE_WORKFLOWS (unchanged); check-ci-filter-parity (reads ci.yml only, which this PR does not touch); the aggregator roster (no ci.yml aggregator involved); merge-queue-triage.yml (workflows: [CI], unchanged). The one reader added is the red-run card filer.
• Install retry (card item 4), not done here: the card's landing spot scripts/ci-setup-pnpm.sh is now ui/scripts/ci-setup-pnpm.sh, and no workflow GitHub runs calls it (only ui/.github/workflows/*); it also never ran the install, only Corepack. The root's pnpm acquisition (.github/actions/setup-pnpm) already retries. pnpm install --frozen-lockfile is spelled at ~40 sites in 25 root workflows. Actions API, failed runs created since 2026-10-19 (632 runs): exactly 1 job failed at an install step, run 39316437234 Type Check · debt ledger, node-gyp headers download ECONNRESET from nodejs.org. The choice of landing (a shared install action at every site, an inline retry, or no change) goes to the devx seat on the card.

Verification

• node scripts/ci/scheduled-full-run.mjs --self-test: OK (battery 46 cases).
• Both changed workflows parse; the new one declares exactly the two jobs and the two triggers above.
• The derived dispatch gates and the three ui pins: see the report on ci: the hourly schedule reruns the full battery on an already-tested SHA, the coverage lane runs on every push, and two steps make unrelated PRs red #25924 (local install is slow in this container; CI's Lint & Repo Gates and Test UI run them on this PR too).
• The new workflow cannot run on a pull request (schedule and dispatch only). The first real reading is a workflow_dispatch on main after merge, or the first night.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WKpunbiE8c9w2i9FsEvQas

[objectstack-fleet]

席位核对

Verdict: ACCEPT at head 565a1e5c40ff2ec9bfeefebbe0c84e97ed0ad01f. This is Tier H: check-governed-merges --pr 26003 reports 2 of 8 paths governed (.github/workflows/**). It lands after a GOVERNED_APPROVERS approval (hotlong / os-zhuang requested), and then the seat flips it.

The premise moved, and the change follows the measured reading

• The card asked to take the coverage lane off every push. After the ui/ cutover the lane runs nowhere: it was row 7 "kept, follow-up port" of the tooling(ui-wiring): objectstack-side inert wiring for the ui/ workspace — workspace globs, root turbo tasks, path-filtered ui test/typecheck/lint/E2E jobs, spec change compiles the Console; pin gates retire at cutover (parent #25677, 3/6) #25697 roster, and ui/.github/workflows is never executed. So ui's four coverage thresholds had no evaluator at all.
• This PR ports the lane to the root as UI Coverage (ui-coverage-nightly.yml). It runs on schedule (13 6 * * *) plus workflow_dispatch only, the thresholds are unchanged, and it has 8 shards with a 40-minute timeout. That is the card's recommendation, applied to the tree as it is today.
• It does not gate anything. No PR or queue build runs it, and RELEASE_GATE_WORKFLOWS is still CI and Lint & Type Check. scheduled-full-run-card.yml watches it, so a red night files or refreshes nightly full run: red on main (UI Coverage).
• No other card or PR claims this port: I searched open devx issues and open PRs for "coverage", and ci: the hourly schedule reruns the full battery on an already-tested SHA, the coverage lane runs on every push, and two steps make unrelated PRs red #25924 / ci(ui-coverage): run objectui's coverage lane nightly from the root, thresholds unchanged #26003 are the only hits.

Seat's own reading at the head (detached worktree)

• node scripts/ci/scheduled-full-run.mjs --self-test gives self-test OK. The battery floor goes from 42 to 46, and the new checks cover four things:
  • the nightly identity's title and marker;
  • no other cadence in a nightly body;
  • the workflow declares exactly one daily cron under the name CARD_CADENCES holds;
  • the watcher list equals the CARD_CADENCES keys.

Gates at head 565a1e5

• 54 success, 12 skipped, 0 failing.

Item 4 (pnpm install retry): not pursued, closed on measurement

• The dev measured 1 install-step failure in 632 failed runs over 7 days. The ui script the card names has no running caller, and the root has about 40 install sites. A retry costs more than it saves at that rate.
• Seat call: option C. Drop item 4 with this reading. After landing, ci: the hourly schedule reruns the full battery on an already-tested SHA, the coverage lane runs on every push, and two steps make unrelated PRs red #25924 closes with the reading recorded. If the install-step failure rate climbs, the card is re-filed with the new numbers.

---

Generated by Claude Code