The ecosystem film explains the project's design and invites developers to contribute. It does not establish complete protocol conformance, deployment security, or mobile voice qualification. Those remain tasks with their own evidence requirements.
The nine chapters cover the goal, authority boundaries, LETS delegation, checks before actions, structured generative UI, component ownership, open protocols, contributor priorities, and the points bounty workflow. Thirty diagram states reveal the mechanisms alongside the narration. The narration contains no academic biography. It uses the Apple-icon-derived AstralDeep wordmark and the Midnight palette. The UI drawings are schematic examples, not product screenshots.
One example task, summarizing selected files, connects the security chapters. Discovery exposes tools, while a separate authorization decision permits file reading and denies email. The delegation diagram splits an illustrative ten-unit budget into seven retained units and three delegated units, with a shorter child expiry. File selection belongs to host policy and bound arguments; these labels are application actions, not literal LETS scope strings. Receipt verification and durable claiming precede a protected action when LETS enforcement is enabled. This ordering does not guarantee exactly-once external effects.
The UI sequence shows an actual container/input/button vocabulary becoming a checked form, followed by the same input and action arranged for desktop and mobile. Projection owns rendering and adaptation, Deep owns orchestration, and Plane owns Astral application persistence. LETS separately owns its warden and executor replay storage. Protocol connections, voice recovery, and contributor workflows use their own process diagrams. The poster previews the rendered-form sequence.
The original editable slide archive and current storyboard with narration and source pointers are retained in the source repository. The original slide archive predates removal of task reservations; use the current storyboard and public playback assets for the current contribution flow. They stay outside the Pages build allowlist. Public playback assets live in site/assets/video/.
The 137.05-second MP4 uses H.264 video at 1920×1080/24 fps and AAC audio. Narration was synthesized through the configured LLM Factory service with speaches-ai/Kokoro-82M-v1.0-ONNX, af_heart, and 24 kHz mono WAV output. English captions are available as both an embedded subtitle stream and a default same-origin WebVTT track. The web track uses small text with bottom padding. A text transcript provides another way to read the story. No service URL or credential is shipped.
Each utterance was synthesized separately. Caption timing follows its audio boundaries, with longer utterances split into at most two lines using proportionate durations. This is sentence timing, not word-level forced alignment. Nine chapter lengths follow the recorded narration. The current edit removes the obsolete reservation instruction and its diagram from original seconds 129.460916667–134.103583333. Remaining narration keeps the accepted voice. The MP4 is re-encoded with H.264/AAC and embedded captions; WebVTT cue times and the transcript omit the same segment. Reveal times follow sentence boundaries, with additional steps inside the budget/receipt explanation. The final file has a fast-start MP4 layout and is approximately 3.1 MB.
Sources were checked at the review revisions:
- AstralDeep guide and engineering constitution: server-owned UI, normal dispatch, identity, owner isolation, policy, confirmation, and audit provenance.
- LETS constitution: attenuated child leases, finite resource budgets, expiry, and verification with durable receipt claims before protected effects. Deep's LETS enforcement is configuration-dependent.
- Component integration intent, Primitives, Projection, and Plane: independently governed components and their responsibilities.
- Conversational voice and client-local speech: voice goals and cross-client qualification obligations. Specs describe intent; the video invites work rather than claiming every client is complete.
- Owner priorities and approval on 2026-10-01, the published source bounty issues, and contribution rules: MCP/A2A compatibility, mobile voice, reusable parts and recognition points.
When revising the story, retain the specified/implemented/tested/live-verified distinctions. Preserve the chosen voice, keep playback under three minutes, check every frame, and inspect desktop and mobile playback before publishing. When only visuals change, preserve the accepted narration and caption timing. Regenerate captions when the narration changes.