Next gate before recorded-data use

The error-budget matrix is a development diagnostic. It does not choose a smoothing window or declare a real recording scientifically usable.

After reviewing the full matrix:

1. Keep all approximation failures visible. A candidate-specific or outcome-chosen window is not permitted. Define an observable motion/support policy separately.
2. Specify the scalar acceleration observable precisely. Mean packet magnitudes and the magnitude of a summed body vector are different measurements; neither can be treated as an instantaneous exact specific force without an error model.
3. Add known acceleration calibration, quantization, time association and signal averaging controls. Preserve the original integer increments and both clock hypotheses where their physical meaning remains uncertain.
4. Combine trajectory truncation, receiver error, nonlinear response and acceleration uncertainty consistently before a finite recorded-data comparison. A local delta/cubature agreement is not calibrated coverage.
5. Freeze a small motion-selected descriptive diagnostic before any original reads. It must compare all candidates, preserve failed windows and withhold a new scientific model-selection claim. No nonlinear observed fit or new campaign allowance follows automatically from these controls.

The original 53-stretch full-position comparisons and the earlier unresolved horizontal-only sensitivity result remain separate.