Chapter one of Creation, and the chapter that opens the track. Delivered in
the Eve edition on 13 September 2026. Ten inherited slides become a
fourteen-slide chapter under the isis id. The brief is
authoring/isis-chapter.md. Machine-readable
state is in state.json.
What was delivered#
Fourteen slides: ten inherited slides rewritten in place and four new — a cover,
two section dividers and a close. Nothing merges away, nothing is carried and no
manuscript retires. content/07-creation-and-studio.json loses seven slides and
keeps eight for the Atelier and editorial chapter;
content/24-creation-deepdives.json loses three and keeps twelve for Atelier
and Living Scenes.
The chapter also creates the Creation track — its label, its summary and its reading route — which is the first half of the A2 plan item done for this track.
Evidence#
No capture. The creator and operator generation surfaces are what this chapter is about, and a screenshot would show a surface without the decision behind it, which is the whole subject. Every claim below was executed at the pin on this host or read there and said to be read.
-
The whole enforcement matrix, executed here. A complete image measurement — seven proofs, one per required gate — run through
evaluateReleaseGateand thendispatchGuardedGeneration, with the injected generation function counted every time:Case Gate mode Dispatch Provider called everything in order allow dispatched 1 one admission denied allow denied 0 safety 0.1 against the 0.9 floor block denied 0 quality 0.4 against the image floor review denied 0 watermark 0.5 against the floor of 1 block denied 0 a review trigger with no reviewer assigned block denied 0 a measurement the gate cannot read block, gate-error:denied 0 -
A proof is bound to one gate. Every proof must carry
policyRef: "isis-release:<gate-kind>". The first run of the probe carried proofs under a general release policy and blocked despite every number passing, with the safety gate reporting "safety score 0.950 ≥ floor 0.9; proof rejected: … evaluated under policy urn:oshun:policy:isis-release, not isis-release:safety-scan". Rebinding each proof to its own requirement, the same measurement allowed. Nothing about the artefact changed. -
Nine gate kinds, not eight.
CANONICAL_RELEASE_GATE_KINDSexecuted here is safety-scan, provenance-c2pa, watermark-coverage, synthesis-disclosure, quality-aggregate, policy-compliance, rights-and-license, output-shape-valid and human-review-ready. The architecture page's headline sentence says eight and lists eight, omitting synthesis-disclosure from that list while describing it further down the same page. -
Eleven human-review triggers, executed here, where the page enumerates seven before trailing off: regulated-topic, clinical-content, legal-prescription, financial-forecast, minor-likeness, real-person-likeness, cloned-voice, spiritual-prescription, political-campaign, cross-domain-memory-write and high-risk-policy-class.
-
The floors do three different things, executed here. Safety below 0.9 blocks. A quality aggregate below its per-kind floor only reviews. Watermark coverage below its floor blocks — unless the floor is zero, as it is for three-dimensional models, point clouds, documents and data, where the gate is not applicable, a fourth status alongside pass, review and block.
-
Read at the pin. Four tiers frozen as
operator-admin,aaa-creator,curated-creatorandcontemplative, where the product-facing labels name the last one Customer; a surface guard that denies by default with a hard 404 and never issues the softer upgrade action its verdict type allows; three producer registries whose state machines all end in a terminalarchived, and only the model machine carries ascanningstate; a provenance bundle with no top-level prompt, consent, user or tenant fields, which live inactors[]andclaims[]instead; provider resolvers that return null without credentials; and a workflow catalog rendered by two independent implementations with a manifest gate that exits non-zero on drift.
What this chapter's evidence tops out at#
Executing the gate and the dispatcher proves what that code does with the inputs it was given, including that the injected generation function was not called. The source says plainly that the live provider HTTP paths are exercised at deploy time with a real key rather than in continuous integration. So this chapter can demonstrate correct refusal and cannot demonstrate generation, and nothing here shows a member receiving an image.
Checks#
| Check | Result |
|---|---|
| Build | 1,691 slides, 705 sources; freshness check clean |
| Python suite | 490 tests pass |
| Node suite | 508 tests pass |
| Browser sweep | 14 slides × 1440, 1280 with audio chrome, 390 and print — 0 findings |
| 14 pages, exact authored visual text; the chapter has no table | |
| Narration | 14 tracks, 24.5 min; library-wide --check verifies 1,691 tracks |
| Bindings | 0 reviewed units cite this chapter; 0 stale across the library |
The sweep passed on the first attempt with no fixes needed.
Tests updated#
Three, each with the reason beside it:
test_build.py's legacy-diagram specimen wasisis-dispatch, whosediagrams.jsonentry retired with this rewrite. It had already been repointed once for the same reason, so it now picks whichever legacygraphslide without declared geometry the library still ships. Five remain, all in Living Scenes guides the Creation track has not rewritten yet.- The tracked-chapter count went 1,271 → 1,285.
- The rollback guard gained
isisinREWRITTEN_CHAPTER_GUIDES.
Treatment deviations#
- No capture, for the reason the brief gives.
creation-audiences: card grid or stat panel →compare-panel. The product labels and the enforced identifiers are two vocabularies on aligned rows, and the mismatch on one row is the teaching.generation-boundaryandisis-dispatch: step-journey →decision-tree. Each is a verdict with named refusals rather than a sequence of handoffs.isis-registry-lifecycle: prose columns →state-machine. Three registries whose lifecycles differ by one state are a machine, not a list.isis-worked: step-journey →case-study. One measurement run twice is a case.
What this chapter does not establish#
No generation was produced, no provider was called and no credentials were used. Nothing here shows a deployed generation path, a released artefact, or a member receiving anything.