ci: re-measure the i18n runtime chunk budget at 740 KB

[CrysisDeu]

Problem / Motivation

Every PR rebased onto current main fails Bundle Size Gate with:

FAIL assets/t-Cr8LfdHT.js: 740.0 KB exceeds its 740.0 KB budget by 4 B (chunk 't')

The PRs' own diffs do not touch the chunk. #8286 (two workflow prompt bullets and one docs paragraph), #8292's fork-review fix and the nested-submenus fix all show the same red on the same head window (08:40–08:43 UTC), while PRs whose merge ref predates 01:24 UTC pass.

Why it matters

The gate is red on main's own content, so it blocks unrelated PRs and trains people to rerun or override it — the same shape as the 2026-08-27 stale-headroom incident recorded in the file's own comment. Until the ceiling moves, nothing that touches main can go green.

What changed (motivation → approach → change)

Three commits landed on main in a seven-minute window (4e5398dd2 agent-picker roster failure, de19df867 inline crew schedule, 390ac8d2f aws-control shared notice), each adding English strings to the t runtime catalog. Individually each fit under the 740 KB ceiling; together they put the chunk 4 B over. The t budget is documented as "grows a little with every translated string, which is expected and fine; what this ceiling catches is a NEW library or surface" — so the right move is to re-measure and move the ceiling, not to shrink strings.

One-line change in website/scripts/check-bundle-size.mjs: t: 740 * KB → t: 780 * KB, with the measurement note updated to 740 KB on 2026-09-04 and the three contributing merges named in the comment. ~5% headroom, matching the all entry's convention.

Tests

N/A — a budget constant; the gate itself is the test. CI's Bundle Size Gate on this PR verifies the new ceiling holds for the current chunk.

Manual verification

Read the failing job log on #8286 (job 100965157938) for the exact overage; confirmed via gh run view that sibling PRs on the same main window fail the same job and that main's own ci.yml skips the gate (PR-only), so main cannot show the red itself.

Related Issues

no linked issue: main-inherited CI unblocker, no tracked issue.

Checklist

• At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)
• Existing tests pass and new tests added for new functionality
• Self-review completed; code follows project style guidelines
• Documentation updated (if applicable)
• No secrets, credentials, or internal references in the diff

[github-actions]

UX Review (Fable 5) — ✅ PASS

UX-level review of f278a8d9cd3aa864490a7b61556103716318a63d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: PASS

No user-facing surface changed — this is a CI bundle-budget comment and threshold bump in a build script; nothing a user sees or touches.

[UX-REVIEWED] f278a8d

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed f278a8d9cd3aa864490a7b61556103716318a63d — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] f278a8d

Verdict parsed from the review's SHA-scoped output markers for commit f278a8d9cd3aa864490a7b61556103716318a63d.

False positive or not applicable? A repository writer can comment:
/ai-review override fable f278a8d9cd3aa864490a7b61556103716318a63d: <one-sentence reason>

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of f278a8d9cd3aa864490a7b61556103716318a63d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

A measured, convention-following ceiling move for documented catalog growth the entry explicitly deems expected — the right fix, at the right layer.

The diff does exactly what the file's own contract demands: justifies the raise, names the contributing merges, keeps the ~5% headroom convention (740 KB measured → 780 KB), and updates the comment in place. String growth is the one cause this entry's docs say should never trip the gate ("grows a little with every translated string, which is expected and fine"), so moving the ceiling — not shrinking strings — is the root-cause-correct response.

Suggestions

• Second identical stale-headroom red in 8 days (08-27, 09-04); a follow-up that warns (not fails) when a chunk sits within ~1% of its budget would let the ceiling move before main blocks unrelated PRs.

[DESIGN-REVIEWED] f278a8d

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of f278a8d9cd3aa864490a7b61556103716318a63d and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f278a8d

False positive or not applicable? A repository writer can comment:
/ai-review override gpt f278a8d9cd3aa864490a7b61556103716318a63d: <one-sentence reason>

[NicholasRBowers]

Tier 1 auto-approve: chore/build tooling (1 files). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, SAST annotations clean, security checklist all-NO, AI reviewers green. Category: single-file CI bundle-budget re-measure in website/scripts/check-bundle-size.mjs (t chunk ceiling 740->780 KB), no runtime impact. Note: edits the same budget line as #8412; whichever merges first conflicts the other, which branch protection resolves.

[CrysisDeu]

Superseded by #8412 (merged 434ff7c), which re-measured the same t ceiling to 777 KB. Closing as redundant.