osmose.calibration.uq.run

End-to-end surrogate-Bayesian UQ orchestrator.

run_surrogate_bayes composes the layer: grow_until_calibrated (Phase 1) -> fit_emulators + make_log_posterior (Phase 2a) -> DynestySampler (Phase 2b). It takes the same injectable evaluator the layer uses (default: the real make_engine_evaluator), so the pipeline is testable synthetically; real engine runs are production usage. Cheap-and-fatal misconfiguration (over-dimension, malformed targets) raises BEFORE the expensive design; the gate-failed case short-circuits without sampling.

Functions

derive_sigma_seed_sq(design, target_keys, ...)	Per-key pooled seed variance s² = n_seeds * mean(alpha[key]_valid).
run_surrogate_bayes(evaluator, free_params, ...)	Run the full surrogate-Bayesian UQ pipeline; return a UQResult.

Classes

UQResult(status, gate_reports, design, ...)

Bundle of a surrogate-Bayesian UQ run.