PASS: reference artwork displayed with GPU mouth deformation; internal rig ready
PASS: idle artwork changes without API connection or microphone activation
PASS: both edge-hover panels open, close on leaving, and do not resize artwork
PASS: invalid-key message, secret cleared, microphone released
PASS: actual Web Audio → HeadAudio → mouth deformation (peak 0.750)
PASS: microphone record → decode → replay → local lipsync (Chromium synthetic microphone)
PASS: camera preview/permission, bounded JPEG over WebRTC, isolated observation, no-person silence, feature greeting, repeat suppression, late-result discard, camera track release
PASS: real local WebRTC/SDP/remote audio + Japanese subtitle events; mouth peak 0.750
PASS: microphone mute/unmute, text input, cancel/clear events, session teardown
PASS: expressions control and narrow-screen layout
PASS: no uncaught JS / WebGL shader errors
NOT TESTED: live OpenAI image/audio API, real passerby recognition, Windows hardware camera/microphone/speaker, Japanese conversational accuracy, human-rated lip-sync naturalness.
