# Living Scenes composition and assist: depth package

Continue the complete V1 teaching audit from
`836abf58de0346dbfe7bd1b318400f128b1dbbc7`. The current library has 79 guides,
1,028 unique slides and 500 cited sources. The overview corrections are the
entry point; they do not replace the detailed composition experience.

Two learning outcomes need their own sequences. The composition guide follows
the three-segment rehearsal through source eligibility, arranging, picking,
previewing, accepting a suggestion, validation, promotion, accessibility and
recovery. The assist guide explains the request and suggestion records, grant
and budget boundaries, filtering, refusals, gold-set arithmetic and challenger
decisions. The technique/primitive catalog, full Score Editor and Tara venue
editor remain distinct connected subjects in the full V1 workstream.

Visual thesis: a cream-paper field guide with large Georgia claims, readable
Arial evidence and terracotta relationships. Use actual component crops for
recognition, source-shaped records for identity, diagrams for adjacency and
branches, and exact comparisons where the reader needs all cases. A coherent
breath → sky → lesson example carries through the composition sequence.

Content plan: begin with the member's authoring task and the explicit rehearsal
boundary, then expose the controls and their effects. Teach source kinds, caps,
boundary ownership and returned records before contrasting web/native behavior
and closing with recovery. The separate assist sequence establishes what the
supplied evidence means before interpreting its metrics. Preserve all existing
guides and recordings, and add source-unit assignments only after the complete
relevant handbook unit has reviewed teaching evidence.

Interaction thesis: keep the current slide entrance, source disclosure and
keyboard navigation. Product crops remain readable in static PDF with complete
alternatives. Diagrams must explain the result without animation or hover.

Read the complete compose helper, web component, mobile component and relevant
owning tests, plus the complete customer-composition and AI-assist handbook
sections. Reconcile overstatements about pair ranking, overlap preview, AI
execution, automatic aborts and persisted/renderable Scores. The source pin is
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`; compare every cited implementation
with the current baseline. Inspect direct callers before attributing an
integration to an interface.

Capture the exact web component using the styles imported by its owning layout
in their declared order. Do not invent missing Tailwind rules or style a mock
and label it as the application. The browser harness may isolate mounting and
static serving; record that boundary. Execute the owning web interaction flows
against that declared host and add cases for boundary ownership, caps and
reload. Execute the native owning host test with its existing mocks; a host test
is not native-device acceptance. No application feature code is changed.

Verify new slide semantics, source claims, final standalone/embedded/narrow
Chrome rendering, accessibility, enlarged text, complete PDFs and narration. Use
focused serial checks with resource supervision. Preserve all baseline content
and recordings exactly, publish using normal hooks to `origin main`, and
synchronize the shared checkout. Human comprehension/listening and the full V1
completion audit remain separate requirements; no partial guide count closes the
active objective.

## Delivered review evidence

The final package contains nineteen composition slides and twelve assist slides,
with five actual UI crops, two adjacency diagrams and six decision diagrams.
Narration totals 1,253.820 seconds for composition and 834.868 seconds for
assist (2,088.688 seconds overall). All 31 final PDF pages received agent
optical review, ordered-text checks and word-bound checks. The final browser
review passes 224 checks with 124 captures and four rejected negative controls;
the center checks both complete guides and preserves the 79 earlier guide HTML
files.

The source/capture work passes 24 characterization cases, 25 owning helper
tests, one native host test and three owning web flows. The presentation run's
124 existing tests pass; the seven new assignment/PDF cases pass after
correcting a synthetic fixture's text order. Earlier passed checks were
retained. Coverage was regenerated for the three newly assigned units. Narration
integrity, builder freshness, scoped lint/format checks and local Markdown
targets pass.

The optical pass replaced two tables with adjacency diagrams, removed tight
label wrapping and rewrote two long visible identifiers that overflowed narrow
slides. The exact implementation references remain in source material; spoken
explanations were preserved. The receipt retains interrupted attempts and
separates automated evidence from agent visual review and unperformed human
acceptance. All 1,028 baseline slides and recordings are preserved. Full V1 and
whole-library design completion remain active.
