Atlas
kb://decisions/2026-05-23-demo-storyboard-optionsdraft2026-05-23

Open + watch + believe" — 60-sec demo storyboard options

demostoryboardphase-1live-demo-ready-barpositioningpolsia-template

Demo storyboard options — 2026-05-23

Strategic research output. NAO directed «изучи тему всю и предложи лучше варианты и узнаем сходится ли наш взор» — full-context study + proposed options to align with his vision.

Phase 1 milestone anchor: live-demo-ready-bar — "Phase 1 anchor — open + watch + believe (live demo-readiness)". A stranger opens the live surface, watches Alice for 60 seconds, groks the product.

Option A: «Alice ships code on air»

Audience: vibecoders + American/SV creators (Cursor/Raycast/Linear register).

Frames (60s):

  1. 0–5s cold open — phosphor cursor blinks in vibecode-claude scene, prompt line ⎇ mvp · ~/openalice · alice@blal.de, no music
  2. 5–15s explorer tree fills with real repo, user message in amber «можешь добавить @feature на новые методы?», assistant streams diff hunks line-by-line (+ /// @feature …), VRM avatar pip top-right with lipsync active, tilts head as assistant turn arrives
  3. 15–25s status bar: feat 12 → 13, tool result: ✓ cargo test -p openalice-storage — 8 passed, avatar SPEAKS the result via ElevenLabs TTS: «eight passed, zero failed» — the moment: an agent ran tests + told you the result
  4. 25–35s ?picker=1 overlay slides in 2s showing all 26 scenes, snaps back
  5. 35–45s cut to atlas.blal.pro constellation + heatmap + Mission Control panel: 1,132 features, 57 repos, 134 runtime edges — Polsia-grade "watch the system work" but observatory dialect
  6. 45–55s back to persona, Alice direct-to-camera (VTuber Fullscreen): «I built that. Want to see what's next?», lipsync precise, lo-fi BGM fades under
  7. 55–60s white-on-midnight URL: live.blal.pro/showcase

Proof-points (all live): vibecode-claude, VRM pip + lipsync, ElevenLabs TTS, atlas with real counters, radio BGM, 26-scene picker.

SOTA angle: no other VTuber/AI streamer shows the agent COMMITTING CODE as content. Neuro-sama chats; Alice ships. Diff-block-on-broadcast is uncopyable short-term — requires vibecode + agent-protocol + atlas annotations end-to-end.

Risk: tail-skip bug (8OOO+ ~20% drop on multi-question bundles) could interrupt spoken test result. Must fix before filming.

Option B: «Two Alices walk into a stream»

Audience: both camps. Hook is visual — vibecoders read A2A; boomers/Germans see two characters in real-time conversation, no humans.

Frames (60s):

  1. 0–5s splitscreen-2 — two VRMs (alice.vrm + female.vrm or male.vrm), both lipsync active, overlay GUEST 1 · GUEST 2
  2. 5–15s agents mid-conversation, chat bridge strip with Twitch + Telegram messages scrolling — no human on screen
  3. 15–25s scene flips to split-2x2 — top: 2 agents, bottom-left: agent navigating browser, bottom-right: terminal. Layout change AGENT-DRIVEN (scene/set WS call), viewer watches it happen because an agent decided
  4. 25–35s cut to persona.blal.pro/demo, user types «tell me what you're building», Alice replies streaming markdown + lipsync, emotion expression fires «happy» — real interaction 3-4s latency
  5. 35–45s back to split, guest agent A2A task completes: ✓ POST /a2a/send — task complete in terminal quadrant — one agent assigned a task to another, we watched it close [NEW — milestone #471 still planned]
  6. 45–55s solo Alice (VTuber Fullscreen): «Every character you see here runs on the same stack. One platform. Unlimited cast.»
  7. 55–60s URL live.blal.pro/showcase + secondary: "26 scenes. One agent. Or two."

Proof-points: splitscreen-2, split-2x2, multi-VRM, persona standalone, ElevenLabs, radio BGM, Twitch+Telegram bridge. Frame 5 is [NEW].

SOTA angle: only streaming platform where TWO autonomous AI agents are both hosts simultaneously on a single canvas, pulling from different chat platforms.

Risk: A2A multi-agent guests milestone #471 is planned, not in_progress. Frame 5 requires it. Can be mocked for proof-of-concept but shouldn't be presented as live without real wiring.

Option C: «The lab built her — watch her work»

Audience: investors / technical founders / enterprise — boomer/German fork. Polsia-template observatory-dialect of "raw numbers make it real".

Frames (60s):

  1. 0–8s atlas constellation rotates, 57 repos / 134 edges / 1,132 features CountUp, Mission Control panel bottom-right, jazz-cafe BGM ambient
  2. 8–18s /timeline page — 18 achievements horizontal ribbon, camera scrubs left-to-right: «Storage v2 deployed», «SLO pipeline live», «Dual-brand locked», «Vibecode 3 layouts shipped» — not a pitch deck, history
  3. 18–30s cut to vibecode-claude scene, real session — Russian question, English answer with diff, status bar reads Wave 8OOO — 22K LOC reshaped. Alice narrates: «Last night I refactored 22,000 lines. Tests stayed green.»
  4. 30–40s atlas /graph force graph (Obsidian-style), nodes pulse, edges animate, voiceover: «Every edge you see is a real runtime dependency.»
  5. 40–52s persona VTuber Fullscreen, Alice: «I wrote 22,000 lines last night. I have 557 tests. Want to see what I do next?» (numbers real)
  6. 52–60s Polsia-template title card (midnight bg, parchment serif, no animation): > OpenAliceLabs. > 57 repos. 1,132 features. > One agent. Streaming live. > She built herself. > atlas.blal.pro

Proof-points (all live): atlas cockpit, /timeline, /graph, vibecode-claude, real refactor stat, 558 persona tests, ElevenLabs.

SOTA angle: "the lab is the product." Cockpit/timeline/graph/tests are real instrumentation, not marketing artifacts. Polsia learning #1 (narrative IS the product) in observatory dialect.

Risk: reads impressive to technical/investor audience but may lose streamer/creator viewer who wants fun avatar action. Atlas UI requires 5s acclimation. «She built herself» — needs careful framing to avoid credibility friction.

Recommendation

Ship: Option A («Alice ships code on air»).

Why:

  • Only option staying on live.blal.pro for most runtime (where the milestone says the moment happens)
  • Uses ONLY live features (no [NEW] gates)
  • One screen recording NAO can make alone, today, without waiting on A2A (#471) or a second agent session
  • The moment (test result spoken as diff block appears) is instantly graspable AND genuinely impossible to do with any competitor's stack
  • Strategic-ideas-log entry «live is the demonstrability surface» confirms this surface gravity

Options B + C aren't wrong — they're for different phases:

  • B → after milestone #471 lands (then B becomes the wide-aperture sequel)
  • C → Phase 3 (investor / public-surface launch, observatory dialect)

Next step if NAO agrees with Option A

  1. Fix tail-skip bug (8OOO+) first — dropped turn during spoken test result = demo killer
  2. Record real coding session in vibecode-claude scene, capture the diff + test-result moment as it ACTUALLY happens (don't script — authenticity is the point per Polsia learning #5)
  3. Trim to 60 seconds
  4. The URL card at the end doubles as a build-in-public post: 60s screen record + «she ran the tests, told me they passed, 26 scenes, one stack» — six lines, Polsia template, zero adjectives

Source files studied: kb://decisions/strategic-ideas-log, kb://brand/dual-brand-architecture, kb://products/alice-persona, kb://products/persona-radio-bridge, kb://products/openalice-atlas-cockpit, kb://research/competitor-signal-polsia-2026-05-22, kb://learnings/2026-05-23-streaming-product-modularization-wave, milestones.yml, scenes-catalogue.ts, memory user-vision/3-phase-plan/dual-brand-direction/streaming-product/live-streaming-no-gpu.