Eve library

Chapter 5 · What may be suggested, and what may leave · brief · 13 September 2026

Guide

Revision 1

On this page

The fifth chapter of the Creation track, and its third merge. living-scenes-assist (12 slides) and living-scenes-governance (10) become one thirty-slide chapter under the new id living-scenes-assist-share, per the Phase D rule that merged chapters take a new id and keep every slide id.

One inherited slide does not come across. scenes-promise was listed by both living-scenes-governance and living-scenes-score, and chapter three already rewrote it in the Eve edition as a step-journey — the treatment this chapter's plan asked for. It stays in chapter three; the merged chapter does not teach it twice. Twenty-one inherited slides are rewritten in place and nine are new.

The job#

Two questions, answered by code on opposite sides of a member's decision. What may be suggested? is Compose Assist: a record schema, a validator, a policy filter, a budget signal, a refusal constructor and a gold-set evaluator. Enumerated over all fifteen runtime exports, exactly two non-test files reference any of them, both front-end components, and ten exports are referenced by neither. What may leave? is the sharing and takedown stack: a nine-component shareability matrix, four reduced-shareability forms, a per-component revocation cascade, seven takedown dispositions, an executor, a nightly SLA audit, three provenance signals and a twenty-four-comparison release gate. One BFF route of eighteen endpoints imports the keep, share and revoke helpers, all seven dispositions, the executor and the Ed25519 signer, and through the share helper it reaches the shareability matrix on every share. The fallback list, the revocation cascade, the nightly audit, the release gate, the channel evaluator and the safety detector are called by nothing at all.

Running example. One kept scene — the sky-to-lesson arc of chapter four — whose narration uses a voice licensed for tenant use only. Compose Assist proposes the bridge; the member keeps the scene; the matrix caps its reach at a password link; the voice is retired underneath it and the cascade runs. Everything counted was executed on this host at the pin against the shipped modules.

Three learning promises (the cover)#

  1. A filter is not a verdict — policy shortens a list; only a caller can refuse, and the reason is never in the list.
  2. The strictest component sets reach — and of the nine components, four have no reduced-shareability form at all.
  3. An empty set is not a pass — four counters in this chapter return a passing result from no evidence, and three more fail closed instead.

Three things to remember (the close)#

  1. Ask what was counted. A gold set of twenty correct refusals scores precision 1 and recall 1; a channel with no fixtures passes; an SLA audit with no impacts reports 1.0000.
  2. Ask which component decided. The matrix returns the deciding components by name; the four fallback forms can only relieve four of the nine.
  3. Ask what the call left behind. A failed cascade reports nothing applied after three mutations already landed.

Bridge. The next chapter takes up the consumers of a shared scene — who opens it, on what device, and what the library still owes them.

Sections#

Four sections, plus the cover and the close.

Section Slides
What assist may propose 5
What the helper will not decide 7
Keep is not share 6
When an input goes away 6

Treatment decisions#

  • Nine slides are new: a cover, four section dividers, a close, and three content slides the executed evidence demanded — scenes-fallback-forms (the four reduced forms and the components they cannot relieve), scenes-cascade-execution (the mutation the nightly audit used to assume) and scenes-empty-denominator (the chapter's spine, counted).
  • The governance half changes source. Its nine inherited slides cited V1/features/scene-*.md and nothing else — they taught a specification. Every one now cites and teaches the shipped module beside the feature document: shareability/matrix.ts, shareability/fallbacks.ts, shareability/per-component-revocation.ts, takedown/dispositions.ts, takedown/execution.ts, takedown/nightly-audit.ts, release-gates/eval-gates.ts, the three provenance/ signal modules and apps/oshun/bff/src/routes/living-scenes.ts.
  • Two inherited tables stay two. scenes-takedown keeps a table because a reader needs seven dispositions and their three consequences side by side; living-scenes-governance-failure becomes a compare-panel, because its four rows are one contrast — what the disposition prescribes against what the executor actually did.
  • One legacy graph retires. scenes-keep-share was the last graph layout in a Living Scenes guide; it becomes the chapter's capture-callouts slide.
  • One real capture. ShareSettingsPanel — the panel a member sees once a share exists — rendered from the pin over a declared fixture, showing the scene URL, the iframe snippet and the oEmbed endpoint as three separate surfaces.

Corrections to inherited claims#

  • "Seven structured dispositions rather than one universal delete." True, and the inherited table showed six of the seven: persona-contract-terminated is a distinct kind from voice-profile-retired, with its own banner and its own re-render option. The table now shows all seven.
  • "The Metis assessment lock is evaluated as an additional restriction." The lock is template-scoped. Executed over all five templates, setting metisAssessmentLocked changes nothing for four of them (7, 6, 6 and 9 techniques survive exactly as before) and takes metis-lesson-visualizer to zero. The same is true of the strictest Lilith tone band: it narrows exactly three techniques, and only on arete-living-offering.
  • "A non-safe crisis label takes precedence and yields private-only before normal grant evaluation." True, and it is stronger than the inherited wording suggests: the crisis branch returns before any validation, so a grant set missing seven of its nine components — or carrying a duplicate — is never reported while a crisis label is set.
  • "Visible marks, attestation, audio watermarking and signed credentials." The four signals are three modules and a page. Two of them were measured here rather than described: the audio watermark round-trips at a bit error rate of 0 with the right key and 0.53 with a key one integer away, and the visible short code is not an identifier — 200,000 sequential artifact ids produce 97,807 distinct codes.

What this chapter does not establish#

No media was rendered, watermarked in production, or shared. The audio watermark was embedded into a synthetic sine buffer, not into a scene's audio. No BFF route was started: the endpoint list is read from the route module, and the modules under it were executed directly. The gold-set arithmetic uses the five-case ledger the source module's own characterisation describes — it is an executed illustration of the evaluator, never a measurement of an assistant, because at the pin no assistant runs.