[backnotprop]

Slice 1 of the Inbox thread extract (plan record, not in this repo). The Inbox thread's pure rules move into @plannotator/core unchanged, so a host of the shared thread parts can use the Inbox's own rules. Plannotator's behaviour does not change.

From (packages/inbox)	To (packages/core, new subpath)
folding.ts	inbox-folding.ts (@plannotator/core/inbox-folding)
deliveryLine.ts	inbox-delivery-line.ts (./inbox-delivery-line; inbox-delivery.ts already exists)
openQuestions.ts	inbox-open-questions.ts (./inbox-open-questions)
format.ts	inbox-format.ts (./inbox-format; openQuestions needs its formatCount)

• Only the import specifiers change. @plannotator/core/x becomes ./x, and ./format becomes ./inbox-format. git diff origin/main:packages/inbox/folding.ts :packages/core/inbox-folding.ts (and the same for the other three) shows import lines only. The modules stay browser-safe and import only core.
• The old paths are export * shims, so every importer (60+) and every test (folding.test.ts, deliveryLine.test.ts, format.test.ts, openQuestions.test.tsx, rowDelivery.test.ts …) runs unedited.
• Core gets 4 new exports. files (**/*.ts) already ships them. No version bump: the owner publishes core by hand.
• The inbox- prefix keeps inbox-e2e.yml triggering on packages/core/inbox*.
• Pi: vendor.sh needs nothing. Pi imports none of these modules, and its core vendor list is explicit.
• Docs: packages/ui/HANDOFF.md gets a new section (marked experimental until Workspaces ships on these subpaths). The guide paths in inbox-messages.md, inbox-window.md and inbox-compose.md now point to the new subpaths.

Proof (local, dev builds, sandboxed)

All of this was run on the branch with origin/main 86538b3 (0.28.32) merged in. The merge brought main's reload-hosts change to deliveryLine.ts (#2213: restartNote for Pi, Oh My Pi and OpenCode 2). It is carried into packages/core/inbox-delivery-line.ts unchanged: each of the four core files still differs from main's packages/inbox original only in its import lines. The guide conflict in inbox-messages.md takes main's text, with waitCount pointing at @plannotator/core/inbox-format.

• Parity harness (test(inbox): pixel and DOM parity harness for the thread #2222, with its follow-ups), a dev build of this branch against a dev build of its merge-base 86538b3: 709 screenshots compared at zero pixels and 829 outerHTML and page-error files. The DOM is identical in all 829 files. One screenshot differs, iphone-15-dark/choices-year-later.png, by 12 pixels. All of them are on the single pixel row where the thread's scroller clips a picture tile, a few colour levels apart. That is WebKit raster noise: the same capture differed when main ran against itself, and its DOM matches. A second main-vs-main run was identical (709 and 829, 0 differing).
  • The harness covers: the list's rows; each thread page by page, expanded, with its open questions and Artifacts; tooltips; the Files band; later clocks (3 and 400 days on); the 👍 Undo window; and live delivery (Undelivered, Send now, the restart line).
  • It runs on: desktop light and dark; the pane at 360 px light and dark, and at 320 px; iPhone 15 (WebKit) light and dark; Pixel 7.
• Earlier harness runs (199 screenshots, 239 DOM files) were IDENTICAL against 948fe2f and against e789a3d. The expanded harness (708/828) was IDENTICAL against e789a3d.
• Unit tests only: format.ts's yesterday branch is covered by the unedited format.test.ts.
• Build order review → hook → opencode → pi, against main 86538b3 built the same way:
  • Every built HTML is byte-identical: apps/hook/dist/{index,review,inbox,surface,redline}.html, apps/review/dist/index.html, and the OpenCode and Pi copies.
  • Pi and OpenCode npm pack --dry-run are identical: 306 and 17 files, same sizes, same tarball shasum.
• bun run typecheck: passes. It covers core's own tsconfig (types: [], so no node:), packages/inbox, apps/inbox, packages/ui/tsconfig.strict-consumer.json and Pi's vendor step.
• bun test packages/inbox packages/core plus packages/server/inbox-thread-folding.test.ts: 0 fail.
• DOM list, the CI way (the 3 DOM_TESTS=1 steps of today's test.yml, the third with --isolate): 6 pass, 40 pass, then 2937 pass with 1 skip. 0 fail.
• Inbox e2e on this branch's binary:
  • inbox-reload-hosts, inbox-restart and inbox-send-now, which draw the moved delivery line: 14 passed, 1 skipped.
  • The full test:e2e:inbox before main 0.28.32: 517 passed. Its two timing flakes passed on rerun.

Not merged, and no versions bumped.