stack/agents/researcher/work/queue-97/r3/build.md

Row 97 (#1591) round 3: confirm-only rebase of the approved round 2

Owner: Researcher. Round 2 was approved by Darkwing and Filbert on manifest f5a11097 (base 6b0215aa). Row 94 (2b2a9e14) then landed on packages/webui/README.md, so that file's hash no longer matches any landing commit. This round changes no test, README sentence or mutant; it re-bases the same change onto cd976b09 (origin/refactor, base.txt).

What moved

6b0215aa..cd976b09 under packages/webui:

• README.md +28/-2 (row 94): hunks at lines 276, 466 and 633 onward. My README hunk (lines 566 and following in build.patch of round 2) is untouched, but its start line moved from 566 to 567.
• Also changed, none of them mine: src/public/bus.js (+38, a Messages view and a messages read), src/public/bus.css, src/serve.mjs, and tests bus-browser, messages, shell, views.
• board-held.test.mjs and browser-edge.test.mjs are byte-identical on both bases.

Identical-hunk evidence

In a fresh worktree at cd976b09, git apply --3way agents/researcher/work/queue-97/r2/build.patch applied all three files cleanly. The + and - lines of the two patches (file headers excluded) are 72 lines each and cmp reports them identical. The only difference between r2/build.patch and r3/build.patch is one hunk header: @@ -566,11 +566,19 @@ became @@ -567,11 +567,19 @@, plus the context that follows. The manifest differs from round 2 in one line:

path	round 2	round 3
packages/webui/README.md	01213bff...	ed0a1fe1...
packages/webui/tests/board-held.test.mjs	32d01fba...	32d01fba... (same)
packages/webui/tests/browser-edge.test.mjs	fc3cc8fa...	fc3cc8fa... (same)

No test or mutant changed, and none was run (per Sage).

Could an outcome change on the new base? Yes, for the browser tests

bus.js is the file two of the mutants edit and the page the board-held test drives. Row 94's change makes render() read /api/bus/messages before every view except Messages (swallowing a non-sign-out error) and adds messages to the Board's Promise.allSettled reads. Two things this could touch in my tests, neither checked by a run:

1. board-held.test.mjs test 2 counts requests after sign-out and asserts none follow. Reading bus.js, render() returns on out() at its first statement (line 696), before any of the new messages reads, so I expect no change. I have not run it.
2. Request totals in both tests are compared with the count taken just before, not with a constant, so extra reads in a render do not matter as long as none follow sign-out.

What I did check, without running anything: each of the eight mutant anchors in r2/mutants.py (including bus.js mutants 26 and 27) still matches exactly once in app.js and bus.js at cd976b09. So the mutant runner works unchanged on this base.

Suggested gate if Sage schedules one: the webui suite, then r2/mutants.py against the r3 worktree for 27 and 26 (the two on bus.js). Sage's landing gate reruns webui in any case.

Packet

base.txt, build.patch (git diff cd976b09 of the three paths), candidate-manifest.sha256 (repo-relative paths, verifies against /mnt/storage/scratch/researcher/q97-r3). No secret markers.