# Chapter 4 · Assembled here, reviewed elsewhere · brief · 11 September 2026

The fourth chapter of the _Tara · practise_ track, rewritten in the Eve edition
on 11 September 2026. It follows the plan in `EVE-V1-REDESIGN-TODOS.md` and the
track brief in [tara-practise-track.md](tara-practise-track.md). The receipt is
under `verification/`.

## The job

Chapter three read a sitting once it was running. This chapter goes back to the
records a ritual is made of before anything plays — and the first thing to know
is that "ritual template" names three of them.

The domain assembler builds one record from a moment rule and a handful of
ingredients. The feature builder normalises a different record, with authored
instructions, a schedule and a review lifecycle, and the Lilith tone reviewer
reads only that one. The shared contract validates a third, with canonical
taxonomy and rights. The bounded search the inherited guide ran finds no code
that converts any of them into another, and the one screen in the Studio that
shows an author a ritual shows none of the three: it shows a two-minute box
breath the BFF serves from a constant.

The rules on each record are narrower than their names. An anchor list is a
choice order, so a supplied alternate is dropped. `requiredKinds` enforces that
an ingredient was supplied, not that the emitted step is required. A duration
maximum is a trimming request that can return eight minutes for a maximum of
four. And a passing tone review binds to a fingerprint of the rendered step
speech — which includes the step numbers and excludes the title, although the
title decides which policy the speech is judged under.

**Running example.** One morning, three records. The assembler's _morning
opening stack_: a ten-minute meditation, a three-minute Nisaba passage, a
four-minute journal and a two-minute Nyx cue, nineteen minutes in four steps.
The feature builder's _morning grounding_: a four-minute box breath, a
six-minute meditation and a two-minute journal, twelve minutes, which the tone
reviewer passes at 0.8458 against 0.7. And the Studio's daily ritual: two
minutes, box breath, 4-4-4-4. Nothing turns one of them into another.

## Three learning promises (the cover)

1. **Three records answer to one name**, and a field means something only once
   the record it belongs to is named.
2. **The assembler's rules are weaker than their names** — a choice order, a
   presence check and a trimming request.
3. **A passing review covers the ordered step speech**, and nothing that sits
   around it: not the title, not pace, not consent, not publication.

## Three things to remember (the close)

1. **Name the record before reading the field.** Local, feature and contract
   templates share a subject and no converter.
2. **Compare what came back with what was asked.** Eight minutes for a maximum
   of four; a journal `required: false` under a rule that requires journaling.
3. **A verdict is current for its fingerprint, not for its policy.** A renamed
   template keeps a review its new policy would fail.

**Bridge.** Chapter five — Tara Mentor Presence: how a teacher's presence is
scored, choreographed and delivered, the last chapter of the track.

## Sections

Five sections of six, seven, five, six and seven.

| Section                       | Slides                                                                                              |
| ----------------------------- | --------------------------------------------------------------------------------------------------- |
| One name, three records       | the case, the three planes, the evidence boundary, the Studio's ritual, six moments, one rule       |
| What the assembler decides    | the morning input and output, the local record, the anchor, a derived passage, a refusal, a journal |
| Which limits hold             | the budget, eight minutes, four minutes asked, the step contract, the template contract             |
| The record the reviewer reads | build and create, the feature record, modality, schedule limits, the fingerprint, the context       |
| What a passing review settles | the policy pack, eight axes, two scripts, the gate, consent, the worked case, the claims            |

## Treatment decisions

- **One new slide, `tara-assembly-studio-ritual`, carries the chapter's
  capture.** The Lilith Studio Tara room is the only application surface that
  shows an author a ritual. It is rendered from the pin over the three answers
  the BFF sends when no member is signed in — the harness evaluates the route's
  own constants and fails if the fixture differs — and it shows a fourth thing
  called a ritual. The same capture shows the Today panel reporting no
  recommendation while the response carries one: the page reads
  `recommendedSit`, the BFF sends `recommended`. Nothing in the pinned web app
  links to the page; it is reachable only by its address. The pins sit in empty
  space beside what they name, checked at the size the slide shows them, after a
  first placement covered the label it pointed at.
- **Three tables survive**, one in thirteen slides: the six moment rules, the
  five step invariants with their exact issue paths, and the eight tone axes.
  Each is a set of exact values a reader compares across rows. The other three
  inherited tables became a graph (the template's policy surfaces), a layer
  stack (the tone context's precedence) and a record specimen (the policy pack).
- **The morning output is a timeline**, not the planned step journey: the four
  steps are consecutive minutes of one sitting, not handoffs between actors, and
  the journey's actors are Eve's.
- **The fingerprint slide is a case study of three edits.** Changing an
  instruction invalidates the review, renumbering the steps invalidates it — the
  renderer prints `order + 1`, so a template authored from one reads "Step 2" on
  its first line — and retitling it does not, though the title chooses the
  policy.
- **The tone context is a five-band layer stack.** Six bands — one per branch of
  the resolver — ran 71px past the narration chrome at 1280×720, so the last
  band holds the three remaining outcomes in their source order: soundscape,
  then unguided, then the guided default.
- **The impossible budget is a stat panel**: requested four, returned eight.
- **The worked boundary is a graph with verified and unverified nodes**, so the
  missing joins are drawn as missing rather than described.

## Corrections to the inherited guide

- `tara-assembly-policy-pack` named the breathing pack's teacher-safety and
  voice policies `tara-safe-teacher-boundary-v1` and `tara-breathing-voice-v1`.
  The pinned registry says `coach-practitioner` and `breathwork-coach`.
- The fingerprint slide said the script records each step's "number"; it records
  `order + 1`, which is the step's number only when the author counted from
  zero.
- The tone-context slide listed the signals without saying they are substring
  matches over the slug, title, summary, tags and step fields: "reset" inside
  "Presets" makes a morning breath a transition.

## What this chapter does not establish

No application path. The assembler, the builder, the contract and the reviewer
are pure functions; nothing here saves, publishes, schedules or launches a
ritual, and no converter between the three records was found. No human review:
no editor, tradition authority, rights holder, member or listener took part, and
the tone verdict is a deterministic text heuristic with its pace axis skipped.
The Studio capture shows what the page renders from the BFF's anonymous answers;
it does not establish what a signed-in author sees.
