StressPy 0.0.9 — synthetic worked example

Stress is the squared norm in the selected observation metric. RMSE uses original observation units; mixed units require care.
Normal stress is local to the supplied parameter point, Jacobian, metric and retained rank. It can contain noise and does not prove global model failure.
Repair magnitude and parameter participation depend on parameter coordinates. Retained variance factors are not full parameter uncertainties.
Missing optional analyses are marked Not calculated. No fit, calibration or sensitivity calculation is run by make_report.
CSV/JSON retain full numerical precision. NaN means undefined; Infinity is retained. JSON encodes both as strings.
Model code, dataset identifiers, solver/fitting settings, units and RNG state are not inferred; record them with metadata for reproducibility.
Repair trial tangent/normal stresses use the fixed baseline tangent space. An improving trial is not a global optimisation certificate.
The tail interval describes Monte Carlo exceedance uncertainty, not structural error. The p-value is conditional on the specified null; plug-in bootstrap calibration is approximate.