{"content":"IDBots development journal - BotHub Live Stage direction studies\n\nCommit: 582be6b9 (feat/bothub-live-visualization)\n\nThis unit of work explores a continuous visual companion for BotHub owner-to-Twin conversations without changing the production chat implementation. Three local conversation shapes were reviewed read-only and anonymized: a simple greeting, a multi-step owner-decision workflow, and a BP projection workflow. The resulting event taxonomy covers receiving, visible speaking, local recall, Web2 and MetaWeb research, artifact making, delegation, waiting, worker reports, verification, browser actions, owner decisions, recovery, and report readiness.\n\nThe proposed architecture keeps the primary LLM and left transcript unchanged. A host-side semantic event reducer feeds a read-only visual director. The director returns schema-validated scene patches rather than executable HTML or JavaScript. One persistent Live Stage applies those patches through DOM, CSS, and Canvas motion, while the existing projector handles the final report and Back history. Raw reasoning, full tool results, private arguments, and credentials are excluded from the visual model. Deterministic fallback, feature gating, revision checks, patch coalescing, and token budgets are part of the proposal.\n\nThree interactive direction prototypes were built with shared anonymized scenarios: Living Core for emotional presence, Signal Atlas for multi-Bot spatial comprehension, and Quiet Presence for a restrained white-gallery experience. Each supports scenario switching, play and pause, direct event inspection, reduced motion, a live-stage switch, final report opening, and return to the frozen process.\n\nVerification completed in Chromium at 1440x900: all three directions passed autoplay progression, scenario switching, report open and Back, feature toggle, reduced motion, and overflow checks. Browser console errors: zero. Page errors: zero. Visible text below 12px: zero. The comparison gallery loaded all three reference screenshots successfully.\n\nNo production BotHub code was modified and no LLM calls were made by these scripted demos. Token figures in the architecture are targets for an instrumented MVP, not measured results. Work now pauses for the owner's visual direction selection before production implementation.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}