Ran .claude/worktrees/nervous-newton-134547/fedac/native/macos/scripts/demo.sh with WALTZ=viennese to capture a brand-new 15.77s MKV from the macOS ac-native build. Script: - Injects a "notepat<enter>" keystroke sequence into the prompt - Captures framebuffer → PNG frames + the CoreAudio synth tap → WAV - Generates `say -v Samantha` TTS greeting synced to f==10 of boot - Muxes everything into one MKV with sample-aligned audio Fresh video structure: 0:00 – 2:00 boot animation with "hi @jeffrey, enjoy Los Angeles!" TTS 2:00 – 3:70 prompt piece with live typing "notepat" + autocomplete 3:70 – 5:00 notepat loading 5:00 – 15:77 notepat playing the Viennese waltz drum pattern 15:77 – 17:28 shutdown strobe (simulated, matches C code exactly) Narration retimed for this 17.28s sequence, still constrained to not talk over the boot TTS or the shutdown audio slot. Cue [0:02.5] now lands on the prompt during "notepat" typing, calling it out directly. Audio fix from the previous commit (concat filter + aresample + async) carried forward — no drift between the fresh demo and the shutdown strobe at the boundary. Caption chip + byline updated: sequence now includes "prompt" and "type 'notepat'" steps that are actually visible in the capture. Cachebuster v=5 → v=6.