Packet timing and independent calibration controls

Both preselected recordings were reproduced from their original packet streams. This audit separates packet timestamp behavior from the unknown integration period of the recorded increments. It identifies neither recording's physically correct clock. Five small independently generated physical controls passed. There were zero observed fit starts and zero native navigation predictions.

An IMU (inertial measurement unit) measures turning and acceleration. These two recordings use IMU type 21 at declared rate code 2 (200 Hz), with the same previously selected intervals. Their source hashes, native packet counts and first/last offsets match the original fit provenance. Every source frame was checked for framing, declared length and checksum. No timestamps or measurements were repaired or interpolated. Group1 fused navigation was not used in this audit.

Observed timing

The duration column is the public receiver-selected stretch. Native counts include its frozen support guard. Mean period error compares header elapsed time with the nominal period, not with an independently measured sensor integration period.

Both streams carry time-types code 2. The public Applanix V6 ICD, Table 3 and section 3.2.3 defines the primary tag as UTC and secondary type zero as POS time, elapsed since power-on. Both tags describe validity time. Their differences provide a useful clock comparison, although independent clock generation or the increment period is not established.

The roughly 22 ppm difference is about 0.0022%, around 900 times smaller than the 2% Z gyro-gain limit. It can matter greatly for accumulated height without directly explaining the repeated large Z gain correction. Clock and calibration must be studied together rather than treating the timestamp difference as the whole explanation.

The report also retains interval extrema, cumulative drift extrema and effective rate. Compatibility with a rounded time lattice is a diagnostic; it does not determine hardware resolution.

Clock bins preserve every consecutive 200-packet block, including partial final blocks, packet offsets, time span and sums of the original six signed integers. These are compact timing checks, not new scientific fit intervals. Original individual measurements remain in the retained source files. Detailed numerical values and input/source/artifact hashes are retained beside this report.

What the independent controls establish

Each timing control generates its increments from a known physical period, independently of its packet tags. Stationary specific force cancels known gravity under the correct integration convention. A known constant-rate angular prediction is checked against the generated angle increments; raw angle increments are not multiplied by dt again. All correct-clock maximum velocity and angular errors are below 10⁻¹⁰ in their units.

The controls show that fixed-period measurements with packet jitter need nominal timing, whereas measurements genuinely accumulated over variable periods need elapsed timing. A timestamp scale error can also create accumulated velocity and height error. These are 120-second one-dimensional physical controls; their errors are not estimates of NASA instrument error.

A further stationary control uses the actual selected UTC timestamp series, but synthetic force increments accumulated over exactly 5 ms. With header timing it produces the height changes below. A single analytically calculated gravity adjustment matches mean duration; it is not a fitted NASA calibration.

This demonstrates a timing/gravity trade-off on the observed time grid. It does not establish that NASA increments span exactly 5 ms, or explain the actual nine-metre vertical discrepancy in full. No observed accelerometer values enter this control.

The actual reoptimized control's gravity correction moved by about +0.040 to +0.042 m/s², while Z accelerometer offset and gain also changed. That is much larger and opposite in sign to the simple stationary mean-duration adjustment above. The saved calibration audit and hashed parameter-change input preserve those values. The single-channel illustration therefore cannot identify the actual compensation path. A coupled three-dimensional control and the matched profile study remain necessary.

For the independent scalar gyro equation, constant rotation leaves gain and offset with rank one. Varying the known rotation raises their rank to two when the clock is known: direct least squares and QR both recover the known predictions and parameters. When clock scale is also free, three parameters still have only rank two. A second gain/offset/clock tuple reproduces the same measured rate exactly to numerical precision. Heading variation can improve gain–offset separation without independently resolving clock scale. This scalar result does not establish the rank of the full GPS/IMU fit; receiver timing and other measured channels may supply additional constraints.

Next test

The reoptimized clock comparison already shows that recalibration repairs most of the fixed-parameter vertical error on the control. The present timing audit and controls do not choose a clock from that better fit. Next inspect accumulated-force and gyro sensitivities under the same calibration, and qualify numerical methods on coupled physical controls before budgeting matched joint Z gain–offset profiles. Use the focused plan. Neither the failed stopping trial nor the original corpus is resumed.

Reproduce this bounded audit:

env PYTHONPATH=analysis OPENBLAS_NUM_THREADS=2 OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 ~/venv/bin/python docs/ilvis0-clock-audit-20261011/build_report.py