Receiver positions give a more useful motion check
The assessment completed all 2,048 preselected centers: 2,046 supported windows and two unsupported edges. The same 1,507 windows retain the earlier airborne classification. No gyroscope measurements or model-fit results were inspected.
Among those airborne windows, 96 position-derived trend intervals lie wholly inside ±0.02°/s even allowing arbitrary temporal correlation. The earlier velocity-derived calculation confirmed none. Under the independent-error assumption the new count is 606; under the supplied correlation sensitivity it is 531. These different counts show why the error assumptions matter.
Point estimates themselves are similar: 756 windows exceed the turn cutoff, versus 749 in the earlier velocity trend. Their median absolute course-rate difference is 0.00129°/s. The main gain is usable uncertainty for some windows, not evidence that the aircraft followed a fundamentally different path.
The median absolute difference between position-derived and mean reported velocity is 0.022 m/s north and 0.013 m/s east. Median cubic residual RMS is 0.0064 m north, 0.0036 m east and 0.0138 m in height. Median residual lag-one correlations are about 0.56–0.60. Polynomial fitting itself affects residual correlation; these values do not identify the true receiver-error law.
What remains open
The 96 intervals are sampled trend checks, not 96 eligible stretches. Centers are spaced apart; this assessment does not establish continuous quiet flight between them. A noise-free fitted trend also differs from instantaneous aircraft motion. No bound on the missing motion was supplied, and the reported receiver covariance remains uncalibrated. Position and velocity are outputs of the same receiver solution, not independent position truth.
The existing ESA inventory still has zero eligible intervals. A replacement selection requires a separately justified and frozen motion/error policy and complete coverage. The seven confirmation groups remain protected.
The run took 43.5 seconds, decoded 64,080 epochs from one receiver file and reconciled all 2,048 window reservations. It performed 2,046 position assessments, point conversions and independent residual fits, within each 2,048-operation ceiling. No original IMU was read and no Earth model was fitted.