ESA: height-dependent gravity prediction

The original twenty-minute recording, earlier/later split, calibration bounds, starts, receiver-error policies and accuracy limits are unchanged. This run changes only the declared gravity law: gravity varies with the predicted height, anchored to the first earlier receiver height. Later receiver heights are opened only after the complete forecast has been saved.

Original pilot · Matched changes for every start.

8/24 starts supplied qualified fits and complete forecasts; 1/4 matched comparisons are complete.

Every setting needs qualified predictions from all three candidates. A missing fit leaves that comparison unresolved; it does not support either geometry. The tables retain every start and physical-accuracy failure.

The one selected interval lasts 20.0 minutes, averages 292.2 km/h near 64.19° N, and contains 360,029 original IMU samples. An IMU is an inertial measurement unit containing gyroscopes and accelerometers. Earlier 800 receiver positions train calibration; later 400 check a continuous IMU forecast. They do not adjust its calibration or path.

RMS describes a typical error size. Height matters alongside horizontal position. Selection uses converged training cost alone, even when another start has a smaller later error. If the selected training fit cannot produce a complete forecast, the comparison retains that failure rather than switching starts. Both starts remain in the full CSV.

These are descriptive comparisons of later total three-dimensional RMS, not calibrated likelihood tests. The shape stage compares the better globe candidate with the disc. Field ranking follows only a complete globe preference whose selected model passes physical accuracy in both sections.

Accuracy requires horizontal RMS ≤10 m, height RMS ≤30 m, maximum horizontal error ≤50 m and maximum absolute height error ≤150 m. A preferred but inaccurate model is not an adequate flight reconstruction. Numerical convergence, physical accuracy and calibrated confidence are separate.

The run used 1,478 native calculations in 2.22 hours. The saved-data audit verifies source and input hashes, exact earlier/later boundaries, every calibration column, repeated source inputs, compiled kernels, physical errors and all charges. It performs no fit or propagation.

The 24 starts and four matched settings reuse one development recording. They are not independent replications. Both receiver-error cases are working assumptions; rate sampling and upstream processing remain explicitly modeled assumptions. No confidence level, field cause or solar-versus-sidereal rate distinction is established here. Seven confirmation groups stay protected, and the original NASA findings remain unchanged.

Every start and its errors · Matched comparisons · Frozen fitting plan · Prepared interval · Roadmap