# Chapter 3 · A scene somebody wrote down · brief · 13 September 2026

The third chapter of the _Creation_ track, and the first merged chapter in it.
`living-scenes-runtime` (16 slides) and `living-scenes-composition` (19) become
one forty-three-slide chapter under the new id `living-scenes-score`, per the
Phase D rule that merged chapters take a new id and keep every slide id. Nothing
merges away: all thirty-five inherited slides are rewritten in place and eight
are new.

## The job

A Living Scene is not a video file. It is a **Score** — an ordered list of
segments, a plan of accepted cues, and a render envelope — that a runtime plays.
The same written thing is validated twice on either side of a process boundary,
steered mid-flight by a closed nine-verb vocabulary that raw text never reaches,
played by a conductor that pre-warms ahead of the playhead, and, when a member
rearranges one, promoted through a gate with a number on it.

**Running example.** One evening scene — a breath anchor, an open sky and a
lesson — scored, validated, played, rearranged and promoted. Executed on this
host against the pinned modules in eight probes: the Score contract and its
runtime mirror run side by side over seventeen malformed scores, the cue parser
over eight hostile strings and twelve parses, the reconciler over seven cue
races, the conductor through its whole pre-warm and advance sequence, the blend
kernel over fifteen pairs and every technique on five templates, and
compose-assist over twenty drafts and promotions.

## Three learning promises (the cover)

1. **A Score is a written contract** — nine top-level fields, eight per segment,
   and a closed cue vocabulary of nine verbs.
2. **Two validators, not one** — executed over seventeen cases, the mirrored
   pair disagreed on thirteen.
3. **Promotion has a number on it** — a continuity floor of 0.85, checked on the
   draft and again at every boundary.

## Three things to remember (the close)

1. **Ask which validator refused.** The runtime checks none of the four identity
   fields; the wire adds bounds the runtime never applies.
2. **Ask which pair is being judged.** A reorder re-pairs every technique
   without re-choosing any.
3. **Ask what the record does not hold.** Six fields, no media, no envelope, and
   the same score id however many times you promote.

**Bridge.** The next chapter takes up the technique catalog and continuity
evaluation in their own right.

## Sections

Six sections of five to seven, plus the cover and the close.

| Section                         | Slides |
| ------------------------------- | ------ |
| What a Score is                 | 6      |
| Steering it while it runs       | 5      |
| Who writes one                  | 5      |
| Arranging a composition         | 6      |
| Choosing what happens at a seam | 6      |
| Promoting, and what it records  | 7      |

## Treatment decisions

- **The chapter keeps its existing captures.** The five `product-view` slides
  become `capture-callouts` over the same five crops, byte-identical, from
  `assets/living-scenes-composition/`. No new capture was taken; the harness
  that produced them is unchanged and its sha256 still matches the recorded
  provenance.
- **Two legacy graphs retire.** `scene-compose-boundaries` and
  `scene-compose-reorder` become `graph-diagram` compositions, closing two of
  the five that remained in Living Scenes guides. Three remain there, in the two
  chapters this track has not reached; nine more sit in other tracks, for twelve
  library-wide.
- **`content/46-living-scenes-composition.json` retires**; it held nothing but
  this chapter's slides. `content/08-living-scenes.json` keeps six governance
  slides and `content/24-creation-deepdives.json` keeps six, both for later
  chapters.
- **Eight slides are new**: a cover, six section dividers and a close.
- **No table in forty-three slides.** Five inherited tables became stat panels,
  compare panels and card grids; the exact values live in the notes.
- **The two worked/failure slides of the runtime guide sit in section three**,
  because they are about the runtime rather than the composition surface. The
  composition surface has its own case and review slides at the end.

## What this chapter does not establish

No frame was rendered, no provider was called, no scene was played and nothing
was promoted into a runtime. Every module executed here is pure: it returns a
decision and performs nothing, and its own doc-comments say so — _"the conductor
models scheduling decisions in deterministic state; actual GPU dispatch lives
outside this module"_. Continuity numbers such as FVD are **inputs** to the
gate, never measured by it. The promotion record carries no media, no render
envelope and no persistence result, and the route from that record into the
standard runtime needs an adapter and execution evidence that does not exist
here — which the chapter states on its own slides rather than only in this
brief.
