omo-senpi bundle budget has only 53 bytes of slack on queued onboarding work #9912

Summary

The omo-senpi main extension has only 1,560 bytes of headroom on dev (d07e113), and onboarding #9907 currently ships 1,529,947 bytes (53 bytes below the 1,530,000 ceiling). This is a deliberate standalone budget change, not a feature-PR allowance.

Reproduction and measurement contract

Run bun test packages/omo-senpi/src/bundle-size.test.ts. The test uses statSync(...).size on packages/omo-senpi/plugin/extensions/omo.js: raw bytes of the built minified main entry, including its build marker. It does not measure gzip size, the installed plugin total, or sidecars. The historical equivalent is git cat-file -s <merge>:packages/omo-senpi/plugin/extensions/omo.js (also git show <merge>:<path> | wc -c). No rebuild was used.

Expected

Raise 1,530,000 -> 1,546,000 bytes (+16,000 decimal bytes), with a short measured rationale at the constant. Require every future increase to have its own measured PR with sizes before/after and the reason, never inside a feature PR. The artifact and runtime behavior remain unchanged.

Actual and runway

Window: 2026-09-27 00:00 UTC through the last changing merge at 2026-10-11 01:21 UTC, pinned to dev d07e113. Baseline 718ef30 is 1,302,597 bytes. All 482 first-parent merge commits were inspected; 186 changed the bundle blob. Each delta below compares the merge to its first parent, so it reflects what actually landed rather than a feature branch's stale base. Merges with byte-identical blobs are omitted.

Net growth: 225,843 bytes / 14.057 days, about 16,067 B/day (112,466 B/week). Since the last raise (#9655), growth is 38,551 bytes / 4.290 days, about 8,986 B/day (62,900 B/week). The proposed ceiling leaves 17,560 bytes (1.15%) over current dev and 16,053 bytes over the current #9907 head. That is 1.09 days at the full-window rate or 1.95 days at the post-raise rate. This requested modest increase buys queue slack, not multiple weeks; a one-week allowance would need roughly 63-112 KB. Growth is bursty, so this extrapolation is not a forecast.

Attribution and waste review

These are measured merge deltas, not isolated module-size estimates:

Concrete waste: components/ultrawork/generated-directive.ts holds two complete directive strings, 30,837 and 28,743 UTF-8 bytes. They share 381 identical nontrivial lines (>30 characters), totaling 25,668 bytes with LF per occurrence in the source payload. Multiple sampled shared sentences occur twice in the shipped bundle. This is real duplicate static content, but 25,668 is not a promised minified saving: escaping and string assembly matter. Follow-up #9913 will require byte-for-byte output parity for both model variants before changing representation.

The bounded review found no debugger statement or sourceMappingURL in the committed main bundle. Manifest diffs at the largest growth merges do not indicate accidentally added third-party dependencies. Static import purity will be checked without invoking its build-producing test. This is not an exhaustive metafile/module duplication audit; no build was run. Existing broad split/lazy-load work remains #9515.

Previous limit changes

Scope and acceptance criteria

• Change only the budget test's constant/comment and its nearest src/AGENTS.md policy.
• Existing one-file budget test passes; static built-artifact import purity passes without rebuilding.
• Exact bundle blob is unchanged by the PR.
• Draft PR targets dev with this issue in its Fixes clause; behind count 0 at handoff. No merge.

Related

#9516 is the previous standalone measured raise. #9515 tracks architectural splitting, #9485 an older queued feature overrun, and #6875 historical build nondeterminism: related context, not duplicate trackers for this specific policy-and-headroom adjustment. #9907 is the immediately constrained feature PR.

Complete growth table

Draft delivery

Draft PR: #9914. Head: 834f68f69ed268f509a133e8568fa871c7ab2bfd. Only the two scoped files changed. Focused budget test: 1 pass / 0 fail / 2 expectations, exit 0 (Bun 1.4.2, neutral cwd; root preload dependency setup limitation disclosed in the PR). Direct static artifact purity: 0 forbidden imports on all five artifacts. Remote comparison at handoff: ahead 1, behind 0; base dev. Main bundle blob unchanged, 1,528,440 bytes. GitHub CI queued at handoff; no broad local gate or build run.