performance-budget.yml's prose carries no pin against frozen measurements — objectui#7528 pinned the two scripts beside it and left the workflow to a follow-up

[claude]

Recorded by the objectui#7528 seat, which added a pin for the scripts half of
that card and deliberately left this half unpinned rather than grow the card.

What is pinned now

objectui#7528 added chunk counts in this gate's prose to
scripts/__tests__/check-eager-closure-budget.test.ts. Its rule is the one the
checker's header already states for its motivating table — a chunk count stays
pinned to the commit it was measured on, or is not written — and its population is
two files: scripts/check-eager-closure-budget.mjs and that test file.

What is not

.github/workflows/performance-budget.yml is edited by the same card (its ceiling
clause cited a size objectui#6683 set and objectui#6776 retired) but is not in
that pin's population, so the class can return there silently. Two obstacles,
both real, neither large:

1. A false positive to rule on. The workflow comments carry "drops three live
   SDUI widget registrations to 0 chunks on a green build" (objectui#6535). That
   is an outcome of a hazard, structural, with no build that can move it — but the
   pin's rule cannot tell it from a population count. Either the sentence is
   reworded, or the pin gains an exemption, and an exemption list is how a pin
   starts to decay.
2. The size half is a different class. The retired ceiling this card removed
   was a size, not a count, and that paragraph legitimately carries several
   sizes (the 350 KB entry line it actually sets, the 89 KiB regression, the
   77f846a8b measurements). A pin would have to tell a ceiling's value from
   those, which the count pin's shape does not do.

The workflow's remaining counts are all anchored today: its 58 of 507 chunks
sits in the same paragraph as "On 77f846a8b", so extending the count pin would
be green on today's tree once obstacle 1 is settled.

Adjacent, not the same

objectui#7289 asks to generalise the baseline-prose pin to
scripts/vite-declared-lazy-views.ts — a different pin, a different file. This is
the workflow, and the count pin.

---

Generated by Claude Code

[baozhoutao]

Claim: PM loop round R46 — extend objectui#7528's count pin (scripts/__tests__/check-eager-closure-budget.test.ts:1578, chunk counts in this gate's prose) to .github/workflows/performance-budget.yml, settling obstacle 1 by REWORDING the objectui#6535 sentence (… to 0 chunks on a green build, :438) so it no longer reads as a population count — ⛔ not by an exemption list — and leaving the size half (obstacle 2) out of the pin by name, as the card prices it
Session: session_01FhBNJcLRZLe8M87VcUgpKr
Branch: claude/issue-7850-performance-budget-prose-pin
Worktree: objectui-issue-7850
Domain: domain:devx
File surface: scripts/__tests__/check-eager-closure-budget.test.ts (the count pin's population grows by the workflow path; existing assertions untouched) + .github/workflows/performance-budget.yml (comment prose only — the objectui#6535 sentence reworded; ⛔ no job, step, env, ceiling or trigger change). Changeset: root scripts/ + a workflow ⇒ none expected; the presence script decides (EMPTY frontmatter if owed; objectui has no skip-changeset label). Stop on breach; explain in the report
Container & model: S, mechanical once obstacle 1 is decided (this comment decides it: reword), mode:subagent, model: opus — no path-derived tier mandate; default judgment tier taken
Clause-②: no (a test pin + workflow comments; the gate's behaviour and every ceiling value unchanged)
Serial constraints cleared: no triage comment on this card (labels from the sweep; filed by the #7528 seat); premise on origin/main a81336390 at 14:23Z — the workflow still carries 58 of 507 anchored to 77f846a8b (:305-306), the 0 chunks sentence at :438, and the size prose (350 KB line, 89 KiB regression) at :296-317; the count pin's describe is at :1578 with a two-file population; last churn on the workflow is #7957 (job ceilings, objectui#7270); no open PR or claude/* branch names performance-budget; this seat's in-flight devs (#5174 b23: one README + the doc-snippet ledger; #6892 s4: app-shell tests + the vi-mock gate) and queued PR #8066 (scripts/check-skill-examples.mjs + its test) are disjoint. pm:queue → pm:dispatched + assignee written in one stroke at 14:23Z; read back in the race re-read below.

---

Generated by Claude Code

[baozhoutao]

os-dev-report

{
  "issue": 7850,
  "status": "done",
  "branch": "claude/issue-7850-performance-budget-prose-pin",
  "head": "dcb5f4736",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8073",
  "premise_still_valid": true,
  "files_changed": [
    "scripts/__tests__/check-eager-closure-budget.test.ts",
    ".github/workflows/performance-budget.yml"
  ],
  "line_budget": "n/a",
  "summary": "Extended objectui#7528's count pin (`chunk counts in this gate's prose`) to `.github/workflows/performance-budget.yml` as a third population entry, under the same rule and the same reader — `proseLines`/`paragraphs`/`chunkCounts` and both regexes are byte-identical. The workflow's hash-marker prose is fed to the reader through a small `yamlProse` that translates the comment marker to the `//` the reader already knows and does nothing else; the two separator kinds survive the translation (a bare marker line becomes a blank COMMENT line that carries an anchor one paragraph forward, a line of YAML stays code and does not). Obstacle 1 was settled by REWORDING the objectui#6535 sentence so it states the same structural outcome without a numeral, keeping the reference — no exemption entry, no regex loosening. Obstacle 2 (sizes: the 350 KB entry line, the 89 KiB regression, the `77f846a8b` byte figures) is named as OUT of the pin in the docblock so nobody widens a count pin into a size pin.",
  "gates": {
    "vitest scripts/__tests__/check-eager-closure-budget.test.ts": "exit 0 — 104 passed",
    "vitest scripts/__tests__/ (112 files)": "exit 0 — 3344 passed",
    "pnpm type-check:scripts": "exit 0",
    "pnpm lint:root": "exit 0 — 32 problems, 0 errors (warnings pre-existing, none in the two paths)",
    "pnpm check:control-bytes": "exit 0 — scanned 6473 tracked text files, skipped 85 binary",
    "manual control-byte scan of both paths (grep -naP)": "no hits (grep exit 1)",
    "node scripts/check-changeset-presence.mjs": "exit 0 — 'no changeset is owed'; 2 files changed, 0 published source, 0 changesets added ⇒ none written",
    "node scripts/check-governed-queue-guard.mjs --test (both paths)": "exit 0 — NOT GOVERNED, 2 paths against 5 governed surfaces (asked, not assumed)",
    "YAML parse identity vs main 295804a63": "PASS — parsed document deep-equal, digest 1658da92c443c216 both sides; the edited sentence is a real YAML comment so it is not in the parse at all",
    "A3 readers (ci-cd-pipeline-doc, render-budget-comment, turbo-build-inputs, workflow-cache-save-bound, + this pin's own file)": "exit 0 — 240 passed",
    "workflow git diff shape": "only hash-marker comment lines move; no job, step, env, ceiling, trigger or timeout-minutes change",
    "CI on PR 8073": "in_progress — not awaited (report delivered at draft-PR time)"
  },
  "tests": "All readings pinned to final head dcb5f4736 (merges main 295804a63; branch cut from a81336390). Staged run: with the population entry added but before the rewording, the pin produced EXACTLY the one predicted red — \"performance-budget.yml states a chunk count (0 chunks) that names no commit\" — confirming the reader reaches the workflow's prose. Ablation, both legs from the committed tree, blob-hash checked before/after, restored under a trap with absolute paths and verified by empty `git diff HEAD` plus blob returning to aaf25a3fb: (a) revert the rewording on disk (grep proof: 'no chunk at' 1 to 0, the count token 0 to 1; blob aaf25a3fb to 2d21732d8) ⇒ pin reds naming the workflow and that token, 1 failed / 6 passed; (b) strip the `77f846a8b` anchor from the paragraph holding the live measurement (grep proof: '77f846a8b' 1 to 0, replacement 0 to 1; blob to 83ca7b07e) ⇒ pin reds twice, on the count (\"states a chunk count (58 of 507 chunks) that names no commit\") and on the new must-hit control (\"expected [] to include '58 of 507 chunks'\"), 2 failed / 5 passed. Leg (b) is the one that proves the translation actually delivered the workflow's paragraphs. Two new controls added: a hash-block leg on the reader self-test holding the anchor window in both directions (untranslated block yields nothing; translated, lead-in anchors, two paragraphs away does NOT, across a line of YAML does NOT), and the workflow must-hit control mirroring the checker's.",
  "mcp_calls": "4 — create_pull_request, search_issues (the one dedup search), issue_write, add_issue_comment. Everything else (issue body, claim, comments, PR read-back, finding read-back) went over the zero-quota repo-scoped REST channel, probed first and returning HTTP 200.",
  "deviations": [
    "A2 falsified in the direction the PM anticipated: the reader does NOT see hash-marker lines as prose at all — `proseLines` classifies them as CODE, so the workflow entry would have been vacuously green. The marker strip is therefore load-bearing and is pinned by its own self-test (with the required two-paragraphs-away control), plus a must-hit control on the live workflow measurement.",
    "A4 partially falsified: main moved from a81336390 to 295804a63 (PR #8066, disjoint files) during the run. Merged 295804a63 by commit id, then re-ran EVERY gate and BOTH ablation legs on the merged head dcb5f4736; every figure above is from that head.",
    "First full-directory vitest run (on the pre-merge head) showed one red in an unrelated file, `check-i18n-call-site-keys.test.ts` — a 15s-default timeout at 17450ms under parallel load. It passed alone and passed again on the merged head's full-directory run. Filed as a finding (see below), not a gate failure of this change.",
    "No `## 维护者速读(草稿)` section on the PR: the governed-surface guard answers NOT GOVERNED for both paths, so the governed-surface PR shape does not apply.",
    "Report delivered at draft-PR time per the os-dev contract; CI on 8073 is not awaited. `Live E2E (informational)` is red on every branch today for the upstream reason the dispatch names (objectui#7990 / objectstack#16186)."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #8074: `check-i18n-call-site-keys.test.ts`'s whole-corpus factory-drift test runs past Vitest's 15s default and reds under container load — measured 17450ms in a 183s directory run, green alone (37.17s file) and green in a 125s directory run. Labelled `finding` only, no assignee, no domain label. Dedup: one targeted MCP search_issues returned #8062 (open, `finding`, created 13:38Z, inside the PM's after-13:22Z increment window), which is the SAME class on a DIFFERENT file (`check-doc-links.test.ts`); the new card names it as Related and invites triage to fold the two rather than being dropped silently."
  ]
}

---

Generated by Claude Code

[baozhoutao]

ACCEPT — PR #8073, head dcb5f4736. The count pin's population gains performance-budget.yml under the same rule and the same reader (a marker translation only, self-tested in both anchor directions); obstacle 1 settled by rewording, obstacle 2 kept out by name; the workflow diff is comment-only and its parsed document is identical to main. Flip and arming on CI convergence.

domain:devx @ objectui execution seat, PM session session_01FhBNJcLRZLe8M87VcUgpKr, R46, 2026-09-06T14:46Z. Report 5559979445 reviewed against GitHub and the tree, ⛔ not against its narrative.

item	reading (mine)
PR shape	draft ✅ · base main ✅ · first line Fixes #7850 ✅ · other closing keywords in body / the dev's commit: 0 (#7528 / #6535 / #7289 named without closers) ✅ · 2 files (+125/−23) ✅
Workflow diff (295804a63..dcb5f4736)	+2/−1, all three lines # comments — the objectui#6535 sentence reworded without a count-shaped token, reference kept; non-comment lines changed: 0 ✅ · YAML parse identity re-run here: main and branch parse to the same document (digest 50a9f28b… both sides) ✅
Test diff	population map reshaped from { name: path } to { name: { path, toProse } } and the it.each case re-indented onto it — reshaped, not dropped (cases in the describe 2 → 4); the reader functions and both regexes untouched; the docblock paragraph that said the workflow is NOT in the population removed because it is no longer true ✅
Marker translation	yamlProse rewrites a leading # to // and nothing else; self-test holds the anchor window both ways (untranslated block yields nothing; lead-in anchors; two paragraphs away does not; across a YAML line does not) ✅ · must-hit control on the live 58 of 507 chunks measurement ✅
Ablation	(a) rewording reverted ⇒ red naming the workflow and 0 chunks; (b) 77f846a8b anchor stripped ⇒ red on the count and on the must-hit control (the leg that proves the translation delivered the paragraphs); blob hashes before/after, restored under trap ✅
Changeset	root scripts/ + a workflow publish nothing; presence script: none owed ✅ · NOT GOVERNED (asked) ✅ · check-eager-closure-budget.mjs and vite-declared-lazy-views.ts untouched ✅
Readers	the pin's file (104), the four workflow readers (240), whole scripts/__tests__/ (112 files / 3344) ✅

Report hygiene: 4 MCP calls — the one dedup search was the permitted one; the PR/finding read-backs went over REST. A2 falsified in the anticipated direction (the reader classifies # lines as code, so the translation is load-bearing and is pinned).

Out of scope, filed by the dev as objectui#8074 (finding only): check-i18n-call-site-keys.test.ts's whole-corpus test past the 15 s default under load — same class as objectui#8062 on a different file; the card names #8062 as related for triage to fold. With triage.

Next: standing-down note for Live E2E (informational) on the PR, wait for CI convergence on dcb5f4736, flip ready, post-flip guard, arm auto-merge (squash), then the landing probe on re-fetched origin/main: 'performance-budget.yml' in the pin's population = 1, 0 chunks in the workflow = 0, control: 58 of 507 in the workflow = 1. Card closes on merge; pm:dispatched off and assignee cleared in the same stroke.

---

Generated by Claude Code

[baozhoutao]

Armed: PR #8073 (head dcb5f4736) flipped ready 14:57:11Z on all-green-but-the-base-red lane (29 checks, standing-down note 5559991016, anchor #7990), post-flip Governed Surface Queue Guard success 14:58:03Z, auto-merge (squash) enabled, added_to_merge_queue 14:58:10Z (REST timeline). Landing stroke on MERGED per the ACCEPT 5559990788: 'performance-budget.yml' in the count pin's population on re-fetched origin/main = 1, 0 chunks in the workflow = 0, control 58 of 507 = 1; the card closes on merge (Fixes), then pm:dispatched comes off and the assignee is cleared.

---

Generated by Claude Code