joshruby/nightjar-night-ambient-20260805

nightjar — 2026-08-05 night ambient, NO DRONE (second cricket night)

A deliberate second, independent drone-free negative set: same rig and site as nightjar-night-ambient-20260804, but a different evening's chorus (2026-08-05, ~21:08 MDT). Audio-only, no video, no DJI — matching the first ambient session's convention.

Why this exists

nightjar-night-ambient-20260804's ~15-20 min of crickets was the only insect-negative material available when training/thresholding against night confusers this campaign. Every attempt using it carried the same open risk: does a fix that works on ONE chorus generalise to insect ambience in general, or did it just learn that specific night's spectral signature? This session is the material that answers it — held out from training ENTIRELY on the 2026-08-05/06 mask-detection training run (job_maskdet.py, blackpaper §11.27), used only as an independent evaluation judge once that run completes.

Contents

5 minutes, both arrays, captured back-to-back with arecord (no flight_record.sh, no tracker/video — a lean ambient capture, same as the 08-04 session):

• uma16_<epoch>.wav — UMA-16, 16 ch, 48 kHz, S16_LE (ch9 dead, standing rule)
• sb_<epoch>.wav — SB-POLARIS, 8 ch, 48 kHz, S32_LE (3 live capsules; detect by RMS)

No DJI, no GPS, no truth — every second is a negative by construction.