A shared error source, separate from the Earth candidates

All six predeclared calculations completed using the first 292 receiver positions of one previously studied IMU21 recording. Later GPS positions, gyroscope observations and Earth-model residuals were absent from the assessment. No Earth-model fit was run.

The calculation puts an upper bound on a common multiplier of each declared receiver-error model. It does not measure receiver accuracy. The Gaussian covariance shape is supplied as an assumption; the calculation cannot prove that assumption from one flight.

These are dimensionless scale upper bounds. Each multiplies all standard deviations in its own declared law. A polynomial that describes the aircraft path poorly can inflate the bound, as the large change across reference degrees illustrates. Both the reference and correlation assumptions matter. All six results are retained; none is selected using an Earth-model preference.

The chart is fixed before calculation, including a longitude scale at 60° latitude. It is distinct from the receiver-dependent coordinate metric of the older fits. The bounds cannot simply be substituted into those fits, and no simultaneous observational coverage or calibrated detection follows.

What this adds

A poorly fitting Earth candidate cannot give itself a larger noise allowance through its own residuals. This training-only component is reproducible, uses exact residual arithmetic and preserves all declared covariance alternatives. The preceding software qualification includes independent exact matrix checks.

The new assessment spent six identities and 5,256 scalar innovation epochs, with zero native predictions, optimizer starts or original IMU reads. The completed freeze cannot be restarted or supply unused work to another study.

What remains

The 53 full-position comparisons remain 53 globe preferences, zero flat preferences and zero unresolved. This error-source calculation adds no independent flight, model winner or statistical significance claim.

The main prediction obstacle remains the broad nonlinear calibration ranges. The generated contraction kept the true parameters but left future ranges hundreds of metres wide. Work must preserve joint calibration constraints and reduce the nonlinear remainder without discarding feasible parameters or using withheld observations. Global numerical error, absolute predictive adequacy and untouched confirmation also remain open.

Every calculation and reference residual · Frozen scope · Exact mathematics · Saved withheld-position evidence · Roadmap.