# Capture work and decisions · chapter 3 · verification receipt · 10 September 2026

Scope: the third chapter of the _Use Eve_ track, rewritten in the Eve edition —
two manuscripts merged into one, sixteen slides re-treated and re-voiced, five
chrome slides added, and every binding re-reviewed. The
[brief](../../authoring/eve-work-chapter.md) records what the chapter is for and
why each treatment was chosen, including three departures from the plan;
[state.json](state.json) holds the exact counts, durations and tool results.

## What was checked

- **Build freshness.** `build-eve-oshun.py --check` passes on the delivered
  tree: 1,371 unique slides, 658 sources, 75 guides. The chapter is 21 slides
  and 28.4 minutes of narration.
- **Teaching crosswalk.** `check-eve-capability-teaching.py` passes across 110
  slides. The eight assignments for this guide carry the rewritten titles, and
  every one of their boundaries was re-read against the new slide and still
  holds, so none needed rewriting.
- **Source contracts.** The recovery machine is checked arrow by arrow against
  `intent-machines.ts` at the pinned revision: each transition names itself, and
  each state box declares which machine states it stands for, so the one box
  that stands for both `leased` and `in-progress` cannot hide a transition the
  machine does not permit. The decision machine on `eve-cap-decision-acceptance`
  is the complete pinned machine — four transitions, five states, and no arrow
  from draft to accepted.
- **Narration.** All 21 tracks rendered and verified: `--check --ids` over the
  whole chapter exits 0, which checks the fingerprint, byte count, SHA-256,
  probed duration and a full decode against the authored pause plan. The other
  1,350 tracks are untouched.
- **Browser rendering.** `eve-edition-review.mjs --guide eve-work-decisions`
  sweeps all 21 slides at 1440×900, 1280×720 with audio chrome, 390×844 and in
  print with no findings: no text outside the viewport, no page overflow, no
  narration chrome pushed off screen, and every authored string painted rather
  than merely present in the document. Every slide was also read at 1440 by eye.
- **PDF.** Exported from the same sweep and held to the exact-text contract by
  `check-eve-specimen-pdf.py`: every authored visual field on its own page, in
  order, with headers, takeaways and captions excluded from the body boundary.
  21 pages pass.
- **Suites and lint.** Python 449 passed with 2,717 subtests; Node 443 of 445;
  Ruff and Prettier clean on everything changed.

## What was fixed on the way

- **The state-machine router painted labels over states.** A diagonal transition
  drew its elbow at the midpoint of the two boxes and put its label there, so
  "reject" landed on the Ready box's title. Transitions are now planned before
  any is drawn: a diagonal picks whichever elbow misses the states in its
  corridor, ends sharing a side of a box are spread across it, and return lanes
  are counted per row. Three new validations refuse what the router cannot draw
  legibly. Committed separately, with tests.
- **A long record field name ran into its own value.** The record specimen's
  label column was a fixed 100px, and `consequences` is wider than that. The
  label column is now one shared subgrid track with a 100px floor, so short
  labels are unchanged and a long one gets the room it needs.
- **Two slides overflowed the 1280 viewport** once the audio chrome was added.
  The compare panel and the export sequence were trimmed until both fit.

## What was not checked

- No human reader session and no listening review at normal speed.
- The narration was rendered on this host's CPU rather than the Apple GPU the
  rest of the library used; only the MP3 bytes differ from what that host would
  produce. No new control re-render was taken this time — the September control
  on `eve-meet-close` matched its committed duration to the millisecond.
- No new product capture was taken. A capture harness re-run on this host's
  Chromium produces different bytes from the committed assets, which would break
  the slide pins, so the chapter reuses the September fixture capture.
- `test_v2_rollback_session` still fails. It guards prior teaching against a
  published baseline, and each delivered chapter rewrite trips it. Re-baselining
  that guard is a decision for the user, so it was left alone.
