{"content":"[IDBots dev journal] fix/webgpu-fallback-resume @ de9661e38 — fix: keep the LiveStage orb recoverable and hold WebGPU startup to its budget.\n\nP1-3 (budget bypass): abortIfSignaled only ran AFTER an await settled, so a requestAdapter()/requestDevice() that never resolves (wedged Windows display driver, suspended RDP session) hung the whole renderer promise — the internal WebGL2 tier never ran and the stage sat on the CSS membrane until the outer 20s timer demoted it for good. New raceWithAbort() races both GPU calls against the 4.5s budget signal (shared by LiveStage and Genesis — both templates inline the same liquid-orb.js), so a hung GPU call now rejects with the timeout error and the existing WebGL2 ladder proceeds. Candidate ladder, adapter fallback skip, and clean-canvas ordering untouched (all test-pinned).\n\nP1-4 (one-way terminal fallback): fallbackOrb used to set orbFailed permanently — one WebGPU or WebGL2 loss (Windows TDR driver reset) demoted the stage to the static CSS membrane until page reload; the host only re-patches state, never reloads. Now every fallback arms ONE bounded retry (30s while visible, immediate on next re-show via visibilitychange; pagehide clears), and a mid-session WebGL2 context loss remounts the WebGL2 orb ('WebGL2 orb lost mid-session, remounting') instead of parking on the membrane. Both remount branches flip data-orb-backend to 'loading' so the CSS membrane covers the swap gap instead of a blank canvas interval. Genesis keeps its own Canvas2D net (unchanged).\n\nTests: studioLiveStageVisuals pins extended (raceWithAbort + budget-signal plumbed acquireHardwareDevice, scheduleOrbRecovery, webgl2 remount, orbFailed=false retry); suites studioLiveStageVisuals + studioGenesisStageVisuals 24/24 pass (NO_COLOR=1), node --check clean on stage.js/liquid-orb.js.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}