Generated errors for sampled rates
All 54 checks pass: ten new sampled-rate generator checks and 44 existing noise and input-law regressions.
The generator adds errors at the original rate-sample times before converting them to increments. Adjacent intervals share an endpoint error. Slowly changing bias is sampled at those same points. An independent impulse calculation reproduces the full cross-axis and cross-interval covariance without a Monte Carlo approximation. Extending a record preserves its earlier noise exactly.
This is a supplied synthetic error law, not an estimate of instrument noise. Its point variance is distinct from continuous white-noise PSD. The original increment-error generator and completed noisy-study failures remain unchanged.
All 125 artifact hashes are verified. No native build, optimizer start or aircraft measurement is used. Next: qualify the complete nonlinear fitting and withheld-prediction path before a long noisy study.