joshruby/nightjar-night-ambient-20260804

nightjar — 2026-08-04 night ambient, NO DRONE

A deliberate negative set: same rig, same site, same pointing, same evening as nightjar-flight-20260804-night, with no drone in the air. Loud summer night insect noise (crickets) on both microphone arrays, plus the camera running its night detector against an empty sky.

Why this exists

Every detection and every cue on a drone flight is uninterpretable without knowing the rate at which the same machinery fires on nothing. Two specific claims from the companion night sortie need this control:

1. Acoustic false-cue rate at night. The cue gate ran in shadow mode (CUE=0), so it logged every candidate it would have posted to audio/cue_log.jsonl as cue_shadow without ever slewing the camera. Anything in that log is, by construction, a false alarm. This matters because the deployed presence gate was relaxed that day to a state-dependent threshold (14.6 dB while tracking, 10.0 dB while searching), and the relaxed setting has never had a measured false-alarm rate. Insect chorus is a genuinely hard case — broadband, loud, and spatially diffuse.
2. Optical false-detection rate at night. The night sortie reported 10.0% detection at 50–100 m — the first thing all day to move that number — but it was not established that those detections were the drone rather than a ground light or a star. Any detection in this session is a false positive, so it bounds that directly.

What is in it

audio/uma16_<epoch>.wav   UMA-16, 16 ch, 48 kHz, S16_LE
audio/sb_<epoch>.wav      SB-POLARIS, 8 ch, 48 kHz, S32_LE
audio/cue_log.jsonl       SHADOW cues — what the gate would have fired on. All false.
video_segments.tar        1080p native + 768x432 substream, 60 s MPEG-TS
blackbox/state.jsonl      tracker FSM ~2.9 Hz    blackbox/frames.tar  substream stills

There is no dji/ directory and that is correct — nothing flew.

Gotchas

UMA-16 0-based channel 9 is a dead electronic-floor channel — exclude it, run on 15. SB-POLARIS has 3 live capsules of 8; detect them by RMS (live ~−70 dBFS, dead ~−190) rather than hardcoding. ⚠ The two arrays use different sample widths (UMA S16_LE, SB S32_LE); reading both as int16 scrambles SB channel identity and destroys its spectrum — that bug produced, and later retracted, a long-range claim elsewhere in this series.

Configuration matches the companion night flight: az_offset_deg 301.37, invert_az true, detector.mode: night, cheap cue daemon (SUBHUNT=0), site notch 100–200 Hz, camera parked at pan 178.9 / tilt 45.

Note the site carries a fixed machinery emitter whose line has been measured at ~140–170 Hz and ~245 Hz across this series; the 100–200 Hz notch suppresses the lower one. Its presence is why per-second energy detection reads ~100% at all ranges on drone flights and why the peak-frequency IQR — wide for a throttle-wandering rotor, pinned for a fixed source — is the discriminator to use rather than raw detection rate.