Eve library

Living Scenes · Twelve techniques, and what admits them · receipt · 13 September 2026

Guide

Revision 1

On this page

Chapter four of Creation, and the track's second merged chapter. Delivered in the Eve edition on 13 September 2026. living-scenes-technique-catalog (25 slides) and living-scenes-compatibility (19) become one fifty-two-slide chapter under the new id living-scenes-technique. The brief is authoring/living-scenes-technique-chapter.md. Machine-readable state is in state.json.

What was delivered#

Fifty-two slides: forty-four inherited slides rewritten in place and eight new — a cover, six section dividers and a close. Nothing merges away and every slide id is kept, per the Phase D rule for merged chapters. 10,420 words of notes and 87.7 minutes of narration, across six sections: what a technique is (5), the twelve and what each costs (6), the primitives underneath (5), continuity and what survives (9), what a pair is asked (9), verdicts, templates and gates (10).

Both source manuscripts held nothing but their own guide's slides and retire: content/48-living-scenes-technique-catalog.json and content/49-living-scenes-compatibility.json. The decks of both source guides are removed.

Three legacy graphs retire with this chapter — scene-tech-layers, scene-compat-pipeline and scene-compat-gating-order — each becoming a graph-diagram composition. Counted from the builder's own output afterwards, nine slides still render as legacy graphs library-wide. One is in a Living Scenes guide: scenes-keep-share in living-scenes-governance, which Creation chapter five reaches. The other eight sit in privacy, Telegram, payments, native, Veritas, Metis, operations and V2 guides that Phase D has not rewritten.

Chapter three's receipt said three legacy graphs remained in Living Scenes guides and twelve library-wide. It missed scenes-keep-share: the figures were four and twelve. Three of those four close here.

A deviation, recorded#

The section-divider layout admits two to six sections, and six sections of seven would cover only forty-two of this chapter's forty-four content slides. The validator's cap wins: the last three sections run to nine, nine and ten slides rather than the usual five to seven.

Evidence#

Everything below was executed at the pinned revision 1ceae52c7896a59ac74a4429d410f3218f0b91c6 on this host. No frame was rendered, nothing was measured from media, and no boundary was played.

  • A technique is twelve fields. Read off the frozen catalog: overlap windows from 0 seconds (hard-cut, jump-cut, cross-cut) to 6 (montage, the only entry over 4; l-cut and j-cut share the widest window of any cut at 1 to 4), maxFvd from 0.30 to 0.60, a provenanceTag per entry, and a reducedMotionFallback that collapses twelve techniques onto four targets — one of which, still-frame, is not itself a composition technique.
  • The freeze is one level deep. The catalog object reports frozen; an entry inside it, its templateAllowlist, its kernelComposition, its evalThresholds, TECHNIQUE_IDS and CRISIS_COLLAPSE_ALLOWLIST all report false. A nested threshold was reassigned at runtime and restored. canReshareDeprecatedTechnique returns true only at the current catalog version.
  • Nine primitives, executed at both edges. All nine transition kinds against both ends of every declared range — bridgeSteps [2, 12], morphFrames [1, 48], fadeMs [50, 8000], deltaDb [-18, 18] — producing three refusal codes: out-of-range, missing-required and unknown-kind.
  • One continuity metric is refused, not defaulted. The scorecard adapter derives four of five numbers and will not fabricate the fifth. Its own comment states the seam: "fvd ← a DECLARED-ABSENT model-bound seam: no in-repo FVD model exists, so callers must inject a real value or an explicit policy sentinel — this adapter never fabricates one", and on a broken probe, "an NaN from a broken probe must fail here, not pass a gate."
  • Eight of twelve techniques have a specific gate. Two omissions are defensible — a hard cut has no transition to observe, a dissolve is the general case. Two are not: smash-cut declares a percussive spike and cross-cut declares parallel action, and neither is checked. The result is silence, not a refusal.
  • The compatibility scorer ships no thresholds. It exports no defaults, and at the pin the only file in the repository that supplies a value for compositeThresholdSoft is the scorer's own test. The running pair — the sky segment beside the lesson — scores exactly 0.700 over seven equally weighted dimensions, and is soft-incompatible at 0.60 / 0.80 and compatible at 0.60 / 0.70. Same seven numbers, opposite verdict.
  • Three verdict regions, two of one kind. Two strict less-than comparisons, so equality clears. Both lower regions return soft-incompatible and differ only in which threshold the rationale names — a formatted English string is the only thing distinguishing them. Malformed thresholds return hard-incompatible, the same kind the four policies use, from an entirely different cause.
  • Validation checks presence, never membership. Executed, both sides carrying an audioRole of didgeridoo score 1.00 on that dimension and come back compatible; so does a sixth tone band, a numeric grounding pin, and an audioRole field absent entirely. The Dice coefficient returns 1 for two empty sets, so two segments citing no grounding at all score 1.00 — better than the running example's 0.50 for sharing one pin of two.
  • The tone helper and the admission gate answer different questions. The helper admits 28 of 60 technique-and-band cells, and hard-cut is the only technique admitted in every band. GatingInput has eight fields and none is a tone band, so the gate cannot ask: fade-to-black on Tara is allowed identically with and without toneBand: 'pedagogical' passed alongside, which it silently ignores, while the helper returns false for exactly that pair.
  • A Metis lock overrides the crisis exception. Four gates run in a fixed order, each with its own code. A hard cut on a locked assessment node is refused with themis-assessment-locked, and a locked Metis lesson filters to 0 of 12 — so the crisis trio does not survive it. filterCandidateTechniques keeps survivors in their original order and discards every refusal code.
  • The same policy is written twice and agrees by hand. The blend kernel and Compose Assist hold per-template availability in opposite orientations, plus their own copies of the crisis trio and the operator-curated Arete set. Executed over all sixty template-and-technique combinations the two decisions agree 60 of 60, with and without the operator grant, while the raw arrays differ for Arete. Nothing binds the three constants.
  • The override record is a constructor, not an audit. recordOverride refuses a hard-incompatible verdict, admits only aaa-creator and operator-admin, and requires sixteen characters of rationale after trimming. It also accepts a compatible verdict, a blank artifact id, both segment ids naming the same segment and a NaN timestamp; it freezes only the outer record — reassigning the nested verdict's kind succeeds — and persists nothing. A durable pattern already exists in this repository, in review-persistence's decision lifecycle, which writes through a repository.

Nothing any probe loaded has moved since the pin: every run reports no drift.

Three of my own claims corrected, by a superlative audit#

Before shipping, every ranking and uniqueness claim in the chapter was re-checked by enumerating the whole twelve-entry catalog rather than the subset a probe had printed. Three had to be fixed, all of them mine and all of them already in a built deck:

  • "The widest overlap in the catalog at 1 to 4 seconds (l-cut, j-cut)." montage runs to 6. Restated as the widest window of any cut, with montage named as the exception.
  • "Techniques fade-to-black can replace: 12 of twelve." Nothing in the source supports that; fade-to-black is the reducedMotionFallback for only itself and whip-pan. Restated as the checkable fact: it is the only entry with both requires and forbids empty, on all five templates.
  • "match-action sits in the middle of every column." It is the second-strictest of its six on all five thresholds, always directly behind fade-to-black. Restated.

A fourth was tightened rather than wrong: smash-cut is the loosest of its six on all five thresholds, not four, and dissolve's 0.03 flicker floor is tied with fade-to-black in the other table rather than held alone.

Two inherited claims corrected#

  • The inherited compatibility slide said the two allowlist tables showed sixty agreements. Executed, the agreement is between the two gates — 60 of 60 decisions, with and without the grant — and not between the two tables, whose Arete rows differ because the kernel keeps the grant logic outside the array while Compose Assist keeps its own operator-curated set in code.
  • The inherited slide said the override helper "does not validate artifact, segment or operator identifiers, or require a finite timestamp". Executed, that is confirmed and extends further: a blank artifact id, both segment ids naming the same segment, a negative timestamp and a NaN timestamp all record.

What this chapter does not establish#

No frame was rendered, nothing was measured from media, and no boundary was played. Every module executed here is pure: it returns a decision and performs nothing. Continuity numbers such as FVD are inputs to the gate, never measured by it. Neither the blend kernel's compatibility scorer nor its assertTechniqueAllowed has a non-test caller at the pin, so nothing here establishes that a product surface applies those functions — which the chapter states on its own slides rather than only in this receipt.

One product surface is shown, and it reaches the policy by the other route. A reader may fairly ask whether any product applies these rules. One does, and not through the kernel: ComposeClient builds its boundary-technique picker by running the Compose Assist twin of the admission filter over all twelve ids, so its options are that filter's output. The capture shows a member being offered seven techniques; it does not show the kernel deciding it. A new harness, tools/presentations/tests/technique-admission-capture.mjs, renders that component from the pin over its own built-in draft and reads the options back off the DOM: every picker offers exactly the seven a Tara contemplative arc admits, and smash-cut, whip-pan, jump-cut, cross-cut and montage are absent from the document entirely. Provenance, including the three assertions and the mocked icon boundary, is in capture-provenance.json.

Two limits on that capture, both recorded in its provenance. Tailwind's utility classes are generated at build time and are not reproduced, and no utility stylesheet was invented to stand in for them — so the images show the component's structure and text, not its deployed visual styling. And the crisis collapse is read back from the helper rather than rendered: ComposeClient hardcodes crisisFrameActive to false, so no screenshot here shows the seven-to-three collapse. Nothing was captured from a deployed studio, a session or a route.

Checks#

  • build-eve-oshun.py and --check both clean.
  • eve-edition-review.mjs --guide living-scenes-technique: 52 slides at 1440×900, 1280×720 with audio chrome, 390×844 and print; zero findings and zero page errors after two fixes — a stat panel declared an exampleLabel its layout does not paint, and one evidence string ran past the 390px viewport. Report in browser-report.json.
  • PDF: 52 pages, exact authored visual text in order. Contract in pdf-contract.json.
  • Eleven tables of fifty-two slides, each at six rows or fewer, so none declares a registry density.
  • check-center-coverage.py reports no stale reviewed bindings anywhere in the library, with and without --check.
  • The full presentation Python suite passes (490 tests, 3,452 subtests), and the Node suite passes (508 tests). test_build.py's tracked-slide count moves from 1,343 to 1,395.
  • Narration: all 52 tracks rendered locally with Kokoro af_heart, 87.7 minutes for the chapter. --check verifies all 1,711 library tracks (39.11 hours) and --check --ids over this chapter's 52 both pass.
  • Ruff, ESLint and Prettier clean on every file touched.

Assignment#

assignments/v1-living-scenes-techniques.json — 11 units, 58 evidence entries, every one citing V1/features/composition-cinematography.md. All 58 were re-reviewed against their rewritten slides before any fingerprint moved, then repointed from the two retired guide ids to the merged one. None was added or dropped, and every unit rationale was rewritten to describe what the rewritten slides now teach. Fingerprints are computed over the slide as build-eve-oshun.py's load_content() returns it, not over the manuscript.

The harness that retired#

tools/presentations/tests/living-scenes-techniques-review.mjs named four files that no longer exist, was never in the Node suite's *.test.mjs glob, defaulted its output to a macOS path, and resolves axe-core from a package that does not carry it on this host — so it could not run here before this chapter either. It is removed. The merged chapter is swept by eve-edition-review.mjs at four viewports and by the library-wide presentation-center.mjs and eve-design.mjs, which run the same visual and table readability audits. Its review config, authoring/v1-living-scenes-techniques-review-config.json, is repointed at the merged guide id. The earlier receipt under verification/v1-living-scenes-techniques-2026-09-09/ remains a historical record of a run against the files as they stood at that revision.