What the interval controls establish

The calibration error in the short synthetic controls has an identifiable numerical component. All nine new fits converged and passed their prediction checks. The original three coarse fits were reused without refitting; no NASA recordings were fitted. See the complete comparisons and figure.

With the same trajectory, calibration, receiver timestamps and solver, halving the integration interval reduced Z gyro-offset error to 24.9–25.1% of its original value. Quartering the interval reduced it to 6.07–6.29%. All three true models show approximately the same square-of-interval trend, consistent with finite-interval approximation error amplified by a weak calibration direction.

At the finest interval, adding prescribed roll while keeping the receiver path and calibration unchanged reduced the remaining offset error by another factor of 34–42. Absolute errors were 0.00084–0.00094°/hour, compared with 0.028–0.039°/hour without added roll. This checks geometry separately from interval size. It does not establish how much roll, bias information or preprocessing exists in the NASA data.

The practical consequence is that a very small position residual can conceal substantial calibration error. Solver convergence alone did not catch it; independent known-motion controls did. This is a reason to qualify the measurement/integration model before interpreting calibration boundaries or expanding observed fits.

The finer samples here were independently generated from known continuous rates. Actual archival packets cannot be subdivided to recover unknown motion within them. The next step is to test an independently derived coning/sculling treatment at the original 200 Hz, under explicit increment conventions. See the next integration check. Original studies and evidence remain frozen.