What calibration leaves in the scalar comparison

All 72 implementation controls passed. They compare four constructed motion histories, three generated truths, three comparison models and two instrument orientation histories. They use no flight recordings. Each history lasts ten minutes and has 121 exact epochs; these are not noisy receiver estimates.

Straight motion loses its scalar distinction

The largest uncorrected eastbound magnitude gap was about 0.0301 m/s². After projection against three gain, three offset and one gravity direction, every eastbound residual was below 1.2 × 10⁻¹⁵ m/s². All suggested linear corrections were inside the declared bounds.

That projection is first order. Its particular correction does not necessarily give zero nonlinear error. The separate constant-gain control already supplies an exact nonlinear construction: a small isotropic gain change alone erases the straight-leg scalar gap. Stationary controls correctly have no model distinction before or after projection.

Changing motion retains a local distinction

The changing-direction and climbing controls retain some difference outside the local calibration directions. Their largest projected RMS values are about 0.00434 and 0.000769 m/s² respectively. These are mathematical discrepancies, not noise-normalized evidence or measured sensor performance.

All 24 wrong-model changing-motion cases had suggested minimum-norm linear corrections outside at least one bound. The report therefore leaves their nonlinear-check fields empty. Applying or clipping those corrections would neither preserve the local solution nor establish a valid bounded fit.

All seven calibration columns remain in the calculation. Stationary cases have rank 4 or 6, eastbound cases rank 6, and the changing-motion cases rank 7 at the declared relative tolerance of 10⁻¹⁰. Full local rank is not a statement about global uniqueness, nonlinear model discrimination or robustness to receiver and calibration uncertainty.

What follows

The next gate is a separately bounded nonlinear calibration control, using the same constructed histories and calibration region. It must check known solutions, analytic derivatives, near-boundary cases and several declared starts; preserve incomplete outcomes and test predicted-observation repeatability. Report feasible residuals rather than interpreting an unbounded linear projection as a realizable calibration. No minimum model separation is required for a software check to pass.

Before recorded-data use, quantization, averaging, time association, receiver derivative error and their dependence still need a consistent treatment. Changing-motion support must be selected using receiver observations before examining gyro or model residuals. Neither these controls nor a future scalar ranking assigns a percentage of the original fit separation to gyroscopes.

The completed full-position comparison remains 53 globe / zero flat / zero unresolved. The separate IMU21 horizontal-only sensitivity result remains unresolved. This work changes neither disposition.

Complete matrix · All outcomes · Frozen scope · Calibration-control design.