Scalar calibration with feasibility certificates

Completed 228 requested-start identities in 11.57 seconds, using 4950 counted residual/Jacobian calls. No observed fits, original-file reads or native integrations.

Known-answer qualification: passed. The matched matrix is complete.

Passing endpoints: 228/228. Passing three-start matrix comparisons: 72/72. A passing endpoint alone does not satisfy a repeat comparison.

Algebraic certificates: 147 identities; actual mathematical optimizer starts: 81. Certificate identities are feasible constructions, not independent optimizer convergence.

Every start and the hashed artifacts preserve failures, parameter points, residual histories, costs, stationarity and numerical calls. Three starts do not certify a global optimum.

The same bounds apply to every candidate. There is no required wrong-model separation; an uninformative scalar diagnostic is a valid outcome. Passing these software checks does not qualify observational uncertainty.

The main result remains 53 globe preferences, zero flat preferences and zero unresolved. The separate IMU21 horizontal-only sensitivity result remains unresolved.

Frozen plan and limits · Local calibration controls · Roadmap.