Z calibration on a preselected different recording

This grid checks whether calibration robustness transfers to the earliest recording already predeclared in the replication ledger. Selection did not use model fits. This is a whole-stretch calibration check, separate from the deferred training/prediction replication experiment.

An IMU (inertial measurement unit) measures turning and acceleration. The file is ILVIS0_applanix_57289_610_LVIS_38614965.013, IMU21, September 24, 2015. The saved interval is 12:47:53–12:55:12 UTC, 439 seconds, with median ground speed 793.61 km/h and minimum 788.77 km/h. Original SHA-256: b1b624d9c5d736b411e6faa58b53000f1884d01563615b2681413234fcd48363. Reuse the complete interval and original measurements, without recalculating eligibility.

The plan specifies the same nine Z gain/offset combinations and three models as the first grid, with fixed-zero upstream removal, original integration and packet-header timing. It has its own 27-start, 5,400-sweep ceiling, no retries. The first pass contained unfinished globe fits; their labels remain unchanged.

The sequential launcher starts this grid after the processing-assumption grid publishes, retaining both locks and using one worker/two threads. Preparation, launch and recovery follow that guide. Completed publication only uses analysis/tests/ilvis0_z_profile_followup.py --study replication --package-only with the frozen environment.

Another date checks transfer across recorded conditions; it does not establish an independent physical instrument or verified installation continuity. This recording already appeared in the archive analysis, so it is development evidence, not unseen confirmation. Group1 fused navigation supplies no inertial evidence. The second predeclared recording and further processing cases are outside this allowance. Statistical decisions remain abstentions; field comparisons do not establish its cause.