Satellite measurements recovered from the airborne recordings

The six prepared recordings now have decoded satellite measurements, rather than only GPS position reports. 19,699 observation epochs contain 603,603 signal measurements. All six original hashes, outer packet counts and complete receiver-record sequences match the previous frozen inspection. Thirty-one focused tests pass. All 232 originals remain preserved; no flight-position fit or Earth-model fit ran.

This is a useful step toward checking the flight trajectory independently of the combined GPS/IMU navigation answer. It does not yet establish the accuracy of the trajectory or resolve what corrections were applied to the IMU.

What has been decoded

The two older receiver streams use concise RT17, with observations every tenth of a second. The other four use RT27, with observations every fifth of a second. No missing observation epoch was found between the complete first and last epochs using the declared 1.5-times-median interval diagnostic. This does not make unfinished final receiver packets complete.

An observation epoch is one receiver measurement time. A signal measurement is one tracked signal from one satellite at that time. These counts are repeated measurements within six recordings, not independent flights or independent statistical trials.

The decoder preserves code pseudoranges in metres, carrier phase in the receiver's original cycle/sign convention, signal strength, available Doppler, constellation, antenna, frequency-band/track identifiers, validity flags and slip information. RT27 signed three-byte clock offsets and six-byte phases, absolute/differential ranges, overflow extensions and variable-length flag/block boundaries are checked. Unknown layouts fail rather than being skipped to improve the usable count. The entire encoded survey payload accompanies each exported epoch. Raw RT27 integers remain available. The RT17 concise and RT27 tables provide the structural and engineering-unit references.

GPS week comes from the recorded receiver products, never the computer's current date. All decoded observation dates agree with the previously established embedded dates. There are 2,968 exact-millisecond observation epochs shared with the receiver's earlier position/error-estimate reports. Millisecond association is explicit; no GPS fix is interpolated into a missing observation epoch. Receiver clock fields are preserved, not silently applied as a guessed correction to IMU latency or observation validity time.

What the orbit and angle checks establish

The streams contain 326 GPS ephemeris packets. These describe satellite orbits and clocks. The GPS decoder retains the transmitted values, converts angular semicircles to radians, and records health, issue identifiers and source offsets. Other satellite information, including GLONASS ephemerides and ionosphere/UTC packets, remains preserved opaque for subsequent decoding. Trimble's GPS ephemeris table and the GPS interface specification support this interpretation; RTKLIB's original receiver implementation supplies an additional angular-unit cross-check.

At recorded position epochs, healthy GPS ephemerides within two hours of their stated reference time predict satellite directions. These directions are compared with the receiver's coarse azimuth/elevation fields. This checks orbit units, timing and association under the usual WGS84/GPS navigation assumptions. It supplies no independent Earth-shape evidence and is not a high-precision orbit or positioning result. Signal travel time and atmospheric delays are not part of this coarse sky-angle check.

Of 29,567 comparisons, 29 exceed the diagnostic limits. Every one is GPS satellite 10 in the April 2009 recording, where the receiver reports both angles as zero from 12:06:59 through 12:07:27 UTC. The propagated elevation is roughly 8 degrees and azimuth roughly 224 degrees. Recorded code/phase observations still exist. The discrepancy is preserved with epoch, flags, measurements, packet offsets and payload hashes; it is not repaired, treated as proof of corrupt ranges, or used to discard the flight. The other comparisons lie within the documented diagnostic allowance for coarse angle fields.

The equal-weight geometry calculation has full local rank at all decoded epochs. Median horizontal dilution factors range approximately 0.55–0.93 and vertical factors 0.92–1.84. It gives each constellation a separate clock parameter. These are geometry factors under assumed independent unit range errors, not metre-valued accuracy estimates or a validated covariance matrix. The zero-angle observations remain in this diagnostic; future positioning must use supported ephemeris geometry rather than trust every coarse angle field.

RT17 phase differences agree much better with one Doppler sign convention than its opposite: median absolute rate-minus-Doppler discrepancies are about 0.07–0.08 Hz after flagged slips/unsupported adjacent intervals are excluded from this labeled check. Both sign hypotheses and all original measurements remain reported. RT27 in these files has no loaded Doppler for this comparison; none is manufactured by differentiating phase. This consistency check is not a carrier-phase position solution or sensor noise bound.

Processing flags and incomplete tails

RT27's recorded code/phase smoothing flags are clear in these four streams. That does not establish the absence of all receiver processing. RT17's flag instead concerns filtered pseudorange corrections; it cannot establish code smoothing. The initial v1 audit mislabeled this distinction. Its original outputs/source freeze remain intact; this separate v2 audit uses an explicit unknown smoothing state for RT17 and a separate filtered-correction field. Those correction flags are clear in both RT17 streams.

The two previously identified receiver tails reproduce: October 2010 ends with an unfinished receiver item and survey record, as does September 2017. Their complete prefixes remain decoded. No resynchronization, range repair, interpolation or original deletion occurred. All outer Applanix framing/checksums remain valid.

What comes next

Reconstruct a bounded receiver-only trajectory from the decoded measurements and GPS ephemerides, with explicit satellite-clock, signal-time and atmospheric corrections. The existing ionosphere/UTC packets and multi-frequency observations can support that work. Compare its residuals and stability with the receiver-reported positions/sigmas, preserving uncertainty from satellite geometry, common errors, filtering and temporal correlation. Comparing two solutions from the same receiver is a consistency check, not independent coverage validation. Carrier-phase ambiguities and slip handling need their own support before a precision claim.

In parallel, the IMU6 processing, physical clock, calibration and mounting uncertainties remain scientific prerequisites. No GPS-only result automatically resolves them, makes Applanix fused attitude independent, transfers a passenger-campaign threshold, or establishes a three-model winner. Measured Earth optimizer starts remain zero.

Code: analysis/lll/trimble_observations.py, ilvis0_gnss.py; worker: analysis/tests/ilvis0_gnss_worker.py. Completed data: data/ilvis0-gnss-v2-20261007/. Each file has deterministic gzip CSV/JSONL measurement exports, a hash-bound report/receipt, and a frozen source/input/environment manifest. This evidence directory preserves the compact reports, per-file inventory, flagged angle inventory and source snapshots. It does not duplicate the larger measurement exports. Both completed audits are frozen and must not be resumed with revised sources.