MAX4466 mic array direction at chance level: ADC clipping (4095) on all channels, plus stale per-channel gains after re-trimming
Rule: when a 4-mic loudness array (ESP32 12-bit ADC, max-min per 25 ms window) calibrates at chance (12.5% on 8 directions, ~25% on 4), analyse the recorded trace before touching the algorithm: count frames at 4095 per channel and print the per-clap peak per channel. Evidence: 15 of 24 claps, then 8 of 12 after re-trimming, drove ALL four channels to 4095 for 7-10 frames, so front/right/back/left read identical and direction vanishes; the few unclipped claps carried clear direction (right 708 vs front 149). Causes: MAX4466 trimmer gain too high for a clap at 1 m, claps of uneven loudness, the front mic near the wearer mouth clipping more (138 clipped frames vs 21-32). Fixes: lower all four trimmers equally until a clap at 1 m peaks about 800-2500, use soft even snaps, exclude saturated frames (any channel >= ~3900) in the estimator and return generic if too few remain, make the calibrate tool detect and re-prompt clipped claps. Second trap: gains calibrated at the old trimmer setting (1.388/1.376/1.0/1.708) become wrong after you equalise the mics (quiet baselines 113/117/112/116, measured gains ~1.0), so re-measure or override locally; do not compute gains from clipped claps (the tool refused to save them, correctly). Keep wearer-specific gains/centroids in gitignored local files.