Python tool for aircraft Doppler audio geolocation using historical ADS-B flight data (Source-available)
Targeted trajectory-motion audit protocol
Issue64 identifies four source identities: a6cc10, ae8b78, ae05a8 and c01125. Their selection comes from the prior failed national scale tiles, independently of any recorder position. Freeze the existing v2 index and canonical15-minute Parquet, time1788899400 through1788900300 (2026-09-08 20:30–20:45UTC), and the two original daily archive parts before comparison. Compute source/member/code hashes and retain missing/failed cases. This is a targeted explanation study, not an estimate of whole-archive defect prevalence or localization accuracy.
Enumerate endpoints and all real interior stationary points of squared cubic Hermite speed in normalized interval time. Its derivative is at most cubic. Verify against constant-velocity and closed-form zero-tangent/noncollinear controls, plus independent SciPy spline samples. Floating-point polynomial roots are not a certified mathematical enclosure. Check every observed interval against65 samples, preserve every interval, and flag speeds >=343m/s as outside the declared globally subsonic model assumption. This does not establish that every observer has a nonunique acoustic solution or that the aircraft actually flew at that speed.
Compare these unchanged inputs:
- Canonical observations with the current gap/dedup/velocity rules.
- Exact stored index payloads, including their payload checksums.
- Raw traces normalized in full before applying the same15-minute window and segment-local velocity rules.
- Fresh full-trace segmentation/derivative support, reporting only intervals whose endpoint times both lie inside the comparison window. This distinguishes window-edge derivative inference from persistent internal source problems.
Join native observation timestamps exactly and report unmatched samples, position/altitude differences, missing motion fields and historical zero fills. Report chord speed, endpoint derivatives, continuous maximum, its time and all unsupported intervals. A chord already exceeding343m/s is a position/time problem under the model regardless of Hermite interpolation. A subsonic chord with an interior peak above that threshold needs separate derivative/interpolant examination. Do not tune thresholds or select an interpolator by recorder error.
Keep production code, source data, acoustic scores and coverage decisions fixed. Fresh normalization alone does not repair an observation jump. Hand measured handling requirements to W2/W3; leave issue64 open until production handling and explicit lost-coverage semantics pass verification. The audit uses no audio, recording coordinates, flight-match truth or new worker deployment.
Follow-up ablation after the first run: the height difference is specifically 137.16m, with historical barometric labels versus fresh ellipsoidal-interpolated labels. Test the normalization order by applying the SAME current normalizer to the clipped raw trace and to the full raw trace before clipping. This tests whether outside-window geometric-minus-baro observations explain the difference; it does not assert either interpolation is independently verified physical altitude. Also count historical nonzero components where the fresh source is missing, because a missing heading treated as zero can fabricate a northward component as well as zero eastward velocity. Preserve the first frozen report alongside this follow-up.