# A stop the start button never reads · Tara · continue, chapter 1 · verification receipt · 11 September 2026

Scope: the first chapter of the _Tara · continue_ track, and the delivery that
declares the track. Nineteen inherited slides became twenty-three: a cover,
three section dividers, eighteen content slides and a close, in three sections
of six. The six rule instalments became one registry and two comparisons, so
`tara-handoff-rules-trauma`, `-violence` and `-specialist` are merged away into
`tara-handoff-rules-immediate`. The
[brief](../../authoring/tara-lilith-handoff-chapter.md) records what the chapter
is for and why each treatment was chosen; the track brief is
[tara-continue-track.md](../../authoring/tara-continue-track.md);
[state.json](state.json) holds the exact counts, tool results and findings.

## What was checked

- **Build freshness.** `build-eve-oshun.py --check` passes on the delivered
  tree: 1,508 slides and 664 sources. The chapter is 23 slides running 31.5
  minutes, with one table, four diagrams and one capture.
- **Track setup.** `tara-continue` is declared in `catalog-source.json` with its
  five chapters, a matching reading route in the same order, and
  `path = ["Products", "Oshun V1", "Tara · continue"]` on each. Three
  prerequisites were added — scheduling on context and lineage, continuity on
  scheduling, reflection on continuity — so every chapter after the first
  depends on an earlier one of its own track. `tara-companions` stays off the
  track until chapter five merges it into `tara-reflection`, so the strip
  counts five chapters from today; the four chapters still in the field-guide
  edition carry it already. The tracked-slide total in `test_build.py` is now
  814.
- **Executed evidence.** The inherited guide's harness,
  `pinned-tara-selection.mjs`, was re-run on this host: 64 owning tests and 21
  source cases pass, and the five modules are byte-identical at the pin and at
  HEAD ([source-report.json](source-report.json)). The running example and
  every new claim were executed against the same source through a transpile
  loader: the grieving case, the two-phrase detection, the context-only case,
  the unknown type, the response planner with and without a region, the
  coverage helper, and every count over the thirteen rules.
- **An adversarial source check.** An independent reviewer checked every claim
  in the draft against the five source files and found eleven that were wrong
  or overstated. Each was verified and fixed before the first build: context
  strings are matched against context cues, not phrases; the classifier's flag
  feeds the mood resolution, so the inputs are not resolved independently; the
  mood's reason is generated for most moods; the Report dialog also takes free
  text; the queued start is never read back; the route also answers 503 for an
  unreachable database; and one rule comes back per call, not thirteen.
- **Capture.** The pinned web sit's finish screen, taken at `1ceae52` for the
  practise track, carries three new callouts, each placed in empty space beside
  what it names and checked on the rendered slide.
- **Bindings.** No assignment or teaching file binds a `tara-handoff-*` slide,
  and `check-center-coverage.py --check` reports no stale bindings. The PDF
  depth test that used `tara-handoff-precedence` as its priority-layout
  specimen now uses `psyche-runtime-fallback`, the one priority ladder left,
  with its literal text taken from that deck's actual PDF export; its four
  removal cases and the reordering case still fail as they should.
- **Browser rendering.** `eve-edition-review.mjs --guide tara-lilith-handoff`
  sweeps all 23 slides at 1440×900, 1280×720 with audio chrome, 390×844 and in
  print — 92 measurements, no findings, no page errors.
- **Narration.** Twenty-three tracks rendered on this host's CPU and verified
  with `--check --ids`; the three merged slides' tracks are pruned, and the
  library-wide `--check` verifies every track.
- **PDF.** 23 pages, exact authored visual text on every one, in order, and
  `check-pdf.py` passes, including its table-header floor on the registry.
- **Suites and lint.** Python 479 passed, 0 failed. Node 489 of 489. `ruff
  check` on the touched tests and Prettier on every touched file are clean.

## What was fixed

- **The sequence ran past the chrome.** The first sweep found the play-button
  sequence 52px past the narration chrome at 1440×900 and 71px at 1280×720.
  Labels that wrapped were shortened and one detail folded into its label, so
  each of the six messages is one line; the notes still carry the detail.
- **Two graph labels collided.** Three inputs arriving from one column shared
  an elbow, and two edge labels printed over each other. The classifier now
  sits above the bridge and the text below it, and no two labels meet.
- **The inherited detector description.** It said a rule keeps its first
  matching pattern; it keeps its first matching signal, with every phrase of it
  that matched, and _I feel unreal, not in my body_ gives one detection with
  both.

## What was not checked

- **No application caller and no enforcement.** Nothing in the pinned
  application calls the bridge; no start is held, no audio paused, no resource
  delivered, and no cooldown, check-in, review or follow-up carried out.
- **No clinical claim.** The rules are source policy; nothing here establishes
  clinical approval, current or reachable contacts, or detection accuracy in
  real speech.
- **No live member data.** Every input is declared, and the capture uses
  declared sample text.
