Software qualification of the continuation method

The control suite checks exact/near-boundary answers, repeated starting points, weak or redundant calibration directions, tight budgets and retained failures. Analytic controls compare Jacobian-column and unit numerical scaling. Short physical controls independently generate increments and receiver positions; their validation uses finite differences rather than the archive's automatic-derivative implementation. The worker also evaluates actual native tangents and writes epoch results for each of the three physical models, checking pre-call accounting and the diagnostic reserve.

A deliberately tight, millimetre-whitened finite-difference fixture cannot verify nanometre-scale prediction stability because of floating-point coordinate resolution. That case must remain unresolved even when the optimizer reports success. The ordinary physical solver controls use 0.1 m-scale uncertainties and retain their original physical-answer tolerances. This modifies software fixtures only; the archive's GPS covariance is unchanged. Passing small controls is not qualification of global optima in minute-long recordings.

The report records the test outcome, numerical environment and source/artifact hashes. These controls make no observed-data fits and confer no Earth-model conclusion.