The audio.c startup-time PCM scan was making ac-native crash with SIGSEGV at dynamic linking time (ld-linux offset 0x1416d, exit code 139 before any stderr output). Not worth debugging inline — the goal of the scan (prefer Speakers over Headset on SOF topologies) is better served by the new speaker.mjs probe piece that lets a human press space on each PCM and listen. On the G7 specifically, /proc/asound tells us pcm0p IS "Speakers" anyway, so the legacy hw:0,0-first order was correct. The scan was solving a problem we don't actually have on this board. Also moved the <alsa/asoundlib.h> include from mid-file to the top block of js-bindings.c — best-practice but also a plausible cause of the weird loader crash when the compiler sees ALSA types introduced only after other functions were already compiled. Kept: - SOF-aware 10ms/40ms period sizing (the real audio fix) - system.audio.listPcms / testPcm (piece-driven probing) - speaker.mjs diagnostic piece + SCORE.md hardware-probing docs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>