GPS uncertainty and the recorded measurement chain

The six-file audit recovered 2,968 receiver positions with their own reported error estimates, without using Applanix's fused navigation as the observation. It also reassembled 29,681 checksummed survey-record envelopes. Six original source hashes and full outer frame counts reproduce, and sixteen focused tests pass. No measured Earth-model fit ran, no scientific rule changed, and all 232 originals remain preserved.

What was hiding alongside the GPS sentences

The primary receiver stream contains binary Trimble messages as well as the familiar GPS sentences. These are separate from the Applanix GPS/IMU navigation answer. The Applanix ICD describes Group 10001 as receiver output, Group 4 as time-tagged IMU data, and Group 10002 as direct IMU output. It does not specify the IMU6 processing history. Its modern clock fields identify validity time; they do not fully establish legacy sensor integration timing or physical latency.

The two older receiver configurations carry GPS-only RT17 survey records. The other four carry RT27 satellite-measurement and enhanced-position record envelopes. Each binary packet has its own checksum, in addition to the enclosing Applanix checksum. Satellite records can span several packets. The new decoder checks page order, reply identifiers, flags and lengths before reassembling them. It preserves the measurement payload rather than claiming to have decoded satellite ranges or carrier phases. The format is documented by Trimble.

The stream also contains GSOF position and uncertainty messages. The parser verifies their pages, record lengths and big-endian engineering units. It retains GPS week, actual epoch, coordinates, status flags, initialization counter and reported uncertainty. Unknown record types remain unpromoted. Trimble's position description specifies WGS-84 coordinates and ellipsoidal height; the uncertainty description defines the reported sigma fields.

What the receiver reports

These are median reported values over each recording, in metres. They are the receiver's own error estimates, not independently measured errors or IMU performance specifications.

Every binary GPS week/date reproduces the embedded receiver date. Every decoded position also matches a unique GPS sentence at the same epoch. Horizontal differences are at the sentence's rounding scale, and height differences remain below about one millimetre after adding the reported geoid separation. This verifies units, byte order, timing and product association. Two products from the same receiver cannot independently verify its accuracy.

The east/north cross-field is preserved as reported. Its public table labels it as dimensionless; the code does not silently interpret it as square metres or assemble a complete covariance matrix from it. Cross-epoch correlation, filtering and the accuracy of the reported sigmas still need support. No millimetre software-fixture uncertainty has been substituted for these metre-scale receiver reports.

Strict handling of incomplete and unfamiliar data

The first audit stopped at NAK response bytes in four streams. The revision recognizes the documented ACK/NAK protocol responses explicitly. They are counted, not discarded, and are not accepted scientific measurements. A NAK can mean that a command is unsupported; it does not by itself show that subsequent satellite observations are invalid. Trimble protocol reference

Four streams end on complete receiver boundaries. The October 2010 file ends inside a receiver packet with 44 bytes remaining; the September 2017 file ends inside a packet header with four bytes remaining. Both also have an unfinished final survey record. These boundary failures remain recorded. No packet, sample or receiver page is repaired or interpolated. Complete preceding packets and GPS products remain available for investigation.

All six Applanix streams are complete and checksum-valid. All accepted receiver binary packets and NMEA sentences pass their checksums. Missing or duplicate survey pages are rejected. There were no GSOF decoding failures. These are separate statements: valid Applanix framing does not imply that the final embedded receiver item is complete.

What is still unresolved for the IMU

All six representatives identify as POS AV 510, version 5, IMU6, spanning three firmware/ICD combinations and separate installation epochs. Their Group-4 raw-frame and IMU status fields are zero throughout. That establishes the recorded status values; it does not establish the meaning of every proprietary status bit or absence of corrections.

Logged messages, including unknown internal messages, are inventoried with their complete payload, hash, count and first/last packet offsets. Modern message numbers are not blindly applied to legacy layouts. Existing inferred installation angles and the marginal 26 October 2010 gyro-scale failure remain unchanged.

Both IMU time tags carry nearly the same microsecond interval fluctuations, with no reversals or clock-base changes. This corroborates the recorded time relationship; it does not independently measure the physical integration period. Receiver packet timing differs from receiver position epochs by tens to hundreds of milliseconds. Those differences include serialization and packet association and are not estimates of IMU latency.

The Group-2 navigation uncertainties are preserved as internal fused estimates. They do not provide independent attitude truth. The IMU6 sensor identity, intrinsic drift/noise, factory calibration, coning/sculling, filtering, Earth/transport compensation and precise latency remain unsupported. Published performance for the different IMU8 reference must not be assigned to these IMU6 files.

The next useful work

The GPS uncertainty problem is now more specific: receiver-reported uncertainty exists and has been decoded, while its covariance and coverage need validation. The raw satellite record envelopes provide a possible route to reconstructing a receiver-only solution and checking uncertainty, subject to complete satellite/ephemeris decoding and available data. That path uses the existing .013 originals; it does not require obtaining other file types.

Separately, establish a defensible recorded-IMU processing and calibration envelope. Then test whether the joint IMU/GPS model can separate a given pair under that envelope. Partial globe-versus-disc evidence remains useful, but the present audit supplies no Earth winner or calibrated significance. Measured Earth starts remain zero.

Code: analysis/lll/trimble_receiver.py, ilvis0_processing.py; launcher analysis/tests/ilvis0_processing_worker.py. Completed output: data/ilvis0-processing-v2-20261007/. This directory preserves the compressed full report, CSV GPS uncertainty inventory, source/input/environment manifest and frozen source copies. The first audit's separate evidence remains intact. Do not resume either completed audit or combine revised sources under their manifests.