# Trust, authority and the current runtime · chapter 9 · verification receipt · 10 September 2026

Scope: the third chapter of the _How Eve works_ track, rewritten in the Eve
edition — twenty-five slides moved out of five shared manuscripts into two,
seven chrome slides added, and every binding re-reviewed. The
[brief](../../authoring/eve-trust-chapter.md) records what the chapter is for
and why each treatment was chosen; [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: 1,405 unique slides, 658 sources, 75 guides. The chapter is 27 slides in
  four sections of seven, six, four and four, running 39.6 minutes.
- **Two revisions, kept apart.** Twenty slides read the documented architecture
  at `1ceae52c`; seven cite files that do not exist at that revision and are
  pinned at `6c3186d2` in their own manuscript. The deck carries both source
  snapshots, and each slide's scope line names the one it was read from.
- **Numbers re-derived from source, not transcribed.** Two new tests in
  `test_supplemental_sources.py` read the pinned revision and recompute what the
  two runtime stat panels claim: the attack and control run counts and per-leg
  successes from the retained measurement record, and the authority-dimension
  count, the recorded absence and the empty sandbox grants from the TypeScript
  they cite. A changed constant fails the suite rather than quietly ageing on a
  slide. The third panel's four threat-coverage counts were read from the dated
  audit table by hand and are quoted in the notes beside it.
- **Assignments re-reviewed, not re-pointed.** Five claims across two assignment
  files bind slides this chapter rewrote — policy order, audit and
  non-repudiation, abuse and adversarial cases, capability grants, and budgets
  and kill switches. Each was read against its new slide before the fingerprint
  moved, and each carries a dated note. Two of those reads found a claim the
  rewrite had dropped and put it back: the client-supplied `permittedToolGrants`
  context on the grant contract, and the four kinds of spending a budget covers
  on the impact slide.
- **Browser rendering.** `eve-edition-review.mjs --guide eve-trust` sweeps all
  27 slides at 1440×900, 1280×720 with audio chrome, 390×844 and in print with
  no findings. Two slides needed fixing first: the worked-attack sequence
  overflowed by 97px at 1440 until it lost a message and shortened its labels,
  and the grant specimen overflowed by 13px at 1280 until its field labels
  became single words.
- **Narration.** All 27 tracks rendered and verified: `--check --ids` over the
  chapter exits 0, and the library-wide `--check` verifies all 1405 tracks
  (1832.1 minutes). Merging four inherited slides away left four orphan tracks
  in the manifest, which the whole-library check rejects; their entries and MP3s
  were removed.
- **PDF.** All 27 pages pass the exact-text contract.
- **Suites and lint.** Python 456 passed; Node 443 of 445; Ruff clean; Prettier
  clean on everything changed. No JavaScript changed, so ESLint had nothing new
  to read.

## What this chapter established

- **A chapter may be pinned at two revisions.** The builder rejects a per-slide
  `sourceRevision`, so the dated-runtime section lives in its own manuscript and
  the guide's slide list interleaves the two. That is now the pattern for any
  chapter whose evidence is newer than the library snapshot.
- **A merge may not consume a foundation slide.** `foundation-coverage.json`
  pins the 244 originally narrated slide ids and the build refuses to drop one.
  Two planned merges were reshaped so the surviving id is the foundation one,
  and one planned merge became two slides instead.

## 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.
- No capture was taken. `assets/eve-product/` holds no crop of the reply-detail
  or voice surfaces, and re-running the capture harness on this host's Chromium
  148 against the recorded 152 would make the assets' provenance false — so the
  treatment checkbox requiring one real capture per chapter stays open.
- The threat-coverage numbers are quoted from the dated audit, not recomputed
  from the matrix.
- `test_v2_rollback_session` still fails against its published baseline;
  re-baselining it is a decision for the user.
