Shared error-scale calculation: software qualification passed
All eight generated families passed, covering 8,192 scale assessments in 4.19 seconds. Every family's exact one-sided 99% coverage lower limit exceeds the frozen 0.97 software threshold. The smallest is 0.97912. This is a finite software qualification around nominal 0.99 model coverage, not measured aircraft coverage. All 44 publication artifacts pass their SHA-256 checks.
The reference projection is fixed before seeing its training errors, and the error model is shared across physical candidates. A candidate cannot enlarge its own measurement uncertainty to explain its poor fit. No withheld positions, archive fits or protected observations enter these controls.
Unlike the earlier bounded-mean construction, this separate method does not need a small reference-trajectory remainder. Under the declared Gaussian law, an incorrect deterministic reference mean can inflate the upper scale bound without making it anti-conservative. The deliberately poor-mean control gives a median upper bound about 873 times the actual scale: safe containment can be uninformative. Central controls give median ratios around 1.20.
Results · Derivation and assumptions · Frozen limits.
Remaining work
The covariance shape and Gaussian law are supplied assumptions. Their adequacy for receiver data, effects of eligibility selection, and an appropriate numerical allowance remain separate observational requirements. A reference learned from the same errors, an unknown distribution, missing numerical allowance or zero residual dimension supplies no scale bound.
This result also does not narrow the recording-scale nonlinear bounds. Those remain too wide for useful prediction intervals. Their finite study ended with three complete histories, one partial and two unstarted; it must not restart. New aircraft prediction fits remain blocked while that method gate is open. The independent gyro-prediction companion can proceed as a descriptive analysis with explicit orientation assumptions and no new fits.