Local Level Lab

Airborne measurements. Open methods.

Paired qualification of wider-arithmetic fits

Completed and qualified: all 16 fits passed in 129 native sweeps. Both targets now favor globe geometry under both processing assumptions, with passing controls and perturbed repeats. The combined findings are 32 globe preferences, zero flat and 21 unresolved stretches. Fit results preserve the evidence.

The completed precision comparison found that wider arithmetic passes unchanged stopping probes at the two cost-only failures. The completed trial confirms that improvement across full model comparisons and repeated starts on the same recordings.

The targets are the IMU8 recording on 24 October 2009 and the DMS recording on 7 October 2015. Each has a completed recorder on an overlapping flight leg. Every target receives all three models under both processing assumptions. Their original whole stretches, measured increments, receiver positions, calibration bounds and numerical tolerances remain fixed. Different recorders retain separate calibration.

The finite plan permits 16 starts and 992 native integration sweeps maximum, with no retries and four total numerical threads. Twelve primary fits and two controls allow forty fitting evaluations each; two perturbed repeats allow two hundred each. Each start also permits two diagnostic predictions. The measured ceiling is approximately 7–9 hours, with earlier completion possible.

Both matched controls must pass. Each target needs all six primary fits and its repeat to converge and agree before its numerical preference can be reported. Failed attempts stay recorded; a repeat cannot replace a primary. The September disc endpoint's separate vertical-stability problem is outside this cost-only trial.

Starts and native calls are journaled before execution, results are hashed and written atomically, and an exclusive lock prevents duplicate workers. The detached worker packages the completed evidence automatically. The trial does not establish statistical significance, a global optimum or the background field's physical cause.