# Tara: practice and continuation · Tara · find a practice, chapter 1 · verification receipt · 10 September 2026

Scope: the first chapter of the third Phase D track, and the track's own setup.
Ten inherited slides became seventeen: a cover, three section dividers, twelve
content slides and a close, in three sections of four. One of the twelve — the
V1.0 release map — is shared with Meet Lilith chapter one and stays in that
chapter's manuscript. The [brief](../../authoring/tara-practice-chapter.md)
records what the chapter is for and why each treatment was chosen; the track
brief is
[tara-find-a-practice-track.md](../../authoring/tara-find-a-practice-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. The chapter is 17 slides running 27.0 minutes, with one table in it —
  the duration bands, kept as a table because five bands across four exact
  numbers is what a reader needs side by side.
- **The claims are executed, not asserted.**
  `tools/presentations/tests/pinned-tara-practice.test.mjs` runs the pinned Tara
  contracts at the same revision the slides cite, in nine tests. It establishes
  the five duration bands and their cadences; that a duration between two bands
  has no band at all and the resolver raises rather than returning one; that
  such a duration escapes `TaraPracticeSchema.safeParse` as a `RangeError`
  instead of becoming a validation issue, because `durationMinutes` is a
  positive number rather than a whole one; that a blocked recommendation must
  name real unsatisfied prerequisites while a satisfied one is checked only for
  an empty blocked list; that the program arc bounds each count against the
  total and not against the other; the nine admitted playback rates with the
  snap, the clamp and the non-finite refusal clamping does not rescue; and the
  mood and modality rules the taxonomy slide names.
- **The capture is the room's own arithmetic.**
  `assets/tara-practice/todays-sit.png` is the top of the shipped `TaraRoom`,
  rendered from its pinned component over the room data the BFF's own
  `buildTaraRoom` produces for two completed sittings. The two completions are
  deliberately not the first two, which is all it takes: the builder offers the
  first unfinished sitting — the second on the path — while reporting a session
  index counted from completions, which is three, and the rail follows the
  number. Above them the kicker reads twelve minutes over a five-minute
  practice, because that string is a literal in the component. Forty-five
  repository files are pinned and hashed; the mocks are the four other rooms'
  own controls, the booking card below the crop, and a router with no route
  table. The run asserts the disagreement off the builder rather than describing
  it.
- **Track setup.** `tara-find-a-practice` is declared with five chapters and a
  reading route in the same order, and every chapter's path moves under it.
  `tara-teachers` stays outside the track until chapter five merges it, so the
  position strip does not renumber itself later. `test_build.py` gained a
  generic guard — every declared track has a reading route in chapter order, and
  every chapter after the first depends on an earlier one of its own track — so
  the next track is covered by the same check rather than by a new clause.
- **One pinned unit re-read, not re-pointed.** The playback-rate unit of the
  Tara feature reference is explained by `tara-audio-accessibility`. Its lines
  were read at the pin before the fingerprint moved: they are unchanged, and
  every field, value and worked case they name is still on the slide. What moved
  is the treatment, from a prose column comparison to a card grid whose evidence
  line carries the two refusals separately.
- **Browser rendering.** `eve-edition-review.mjs --guide tara` sweeps all 17
  slides at 1440×900, 1280×720 with audio chrome, 390×844 and in print — 68
  measurements, no spill, no document overflow, no page errors, no findings.
- **Narration.** Sixteen tracks rendered on this host's CPU in batches of six
  and verified; the library is complete at 1903.4 minutes with nothing pending.
  The seventeenth slide is the shared release map, whose text did not change.
- **PDF.** 17 pages, exact authored visual text on every one, in order, read
  back from the downloadable `decks/tara.pdf`.
- **Suites and lint.** Python 473 passed, 0 failed. The node presentation suite
  is green. `ruff check`, ESLint on both new harnesses and the contract test,
  and Prettier on every touched file are clean.

## What this chapter found

- **The room's kicker is a literal.** `Today’s sit · 12 min` is written into the
  component, and the practice under it is whatever the builder offered. On the
  captured data those are twelve and five.
- **A count is not a position.** `sessionIndex` is the number of completed
  sittings plus one; the offer is the first unfinished sitting. They agree only
  while the completions are an unbroken run from the start, and the progress
  rail follows the count.
- **A validator that raises instead of rejecting.** The five duration bands are
  contiguous over whole minutes and nothing else. A practice declared at two and
  a half minutes is field-valid, reaches the bucket resolver inside the
  refinement, and the `RangeError` there escapes `safeParse`. The check that
  exists to catch a mismatched bucket is the one that breaks on it.
- **A readiness flag the schema does not recompute.** The blocked branch of a
  continuation state is verified against the prerequisite list; the satisfied
  branch is verified only for an empty blocked list. A state may therefore claim
  readiness while holding a prerequisite that is not met.

None of the first three is fixed here. They are product and contract work, and
the chapter's job was to find and teach them; each is named on the slide it
belongs to and recorded in [state.json](state.json).

## What was fixed

- Both five-step journeys ran 3.5px past the page at 1280×720 with the
  narration chrome, because three-line step titles push the card grid over.
  The titles are two lines each now.
- The capture slide first ran 40px over at the same viewport with four
  two-line legend entries; each is one line now.
- `check-v2-quality-pdf.py` read a quantitative table cell verbatim, so the
  first table with a declared missing state crashed it with the state key. It
  compares the painted marker now, which is what the reader sees.

## 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
  older part of the library used; only the MP3 bytes differ.
- Nothing about the player. Whether a normalized rate is applied, whether breath
  cues survive a pause, and whether a transcript is present on a given surface
  belong to _Tara · practise_, and the chapter says so on the slide that would
  otherwise be read as covering them.
- Nothing about devices or durable memory. Haptic delivery and rights-preserving
  downloads belong to the native track; cross-device continuation belongs to
  _Tara · continue_.
