What passed and what blocked advancement

This software qualification is not complete enough to permit observed refits. It stopped at the first failed control, after 24 passes. Eighteen planned controls were not run. The original 53 recorded-flight comparisons are unchanged.

Passing checks

The completed checks include exact bounded solutions, retaining an explicit unobservable parameter, invalid inputs, finite evaluation limits, and conversion of retained prediction changes into physical distances. Six native derivative checks retained all 26 or 27 parameter columns and agreed with separate value-only calculations within the declared numerical tolerances.

All six horizontal-only generated-motion fits reached so far passed: each of the three models under nominal and variable intervals, starting at zero. Each matched its independently generated horizontal observations within 2 mm and passed the fresh numerical checks. Their opposed-start checks were not reached, so horizontal-only refitting is not yet qualified either.

Failed check

The first height-only fit used rotating-globe generated motion, nominal intervals and a zero start. Its largest height error was only 0.00061 mm, but its fresh stationarity measure was 0.0001409, above the unchanged 0.0001 limit. It used 79 tangent sweeps within its 80-sweep ceiling and stopped at its fitting allowance. The unused reserve does not authorize a retry.

The same solution's largest error across all coordinates was 10.69 m. That is possible because horizontal coordinates did not enter this objective. The fit therefore illustrates both why a tiny scored residual is insufficient for numerical acceptance and why omitted coordinates need separate reporting.

All six columns had nonzero numerical singular values, but the ratio of largest to smallest was about 26 billion. A full numerical rank does not imply that each calibration parameter is well determined. Deleting weak columns would change the problem; none were removed. Wider internal arithmetic did not remove the development stopping failure, and the sealed batch used that same wider arithmetic with 64-bit mantissa precision.

Next numerical question

Determine which combinations of calibration and timing parameters create the weak directions, and whether a better bounded step can satisfy the existing fresh checks without claiming unique calibration. Any new endpoint calculations or revised qualification need their own finite freeze. Do not restart this completed batch or advance to recorded-flight refits from partial successes.

The batch used seven generated-motion starts and 459 native sweeps, with no original-recording reads or observed-data fits.

Recorded controls and limits · Frozen plan · Measurement-contribution work.