Calibration Workflow

The eight steps examples/calibration_workflow_complete.py runs, in order.

Step	What happens	API
1	Build the plant, then generate a twin record from it	create_twin_measurements
2	Validate, remove outliers, fill gaps	measurements.validate, remove_outliers, fill_gaps
3	Fit, with a train/test split	InitialCalibrator.calibrate
4	Read the metrics of both parts	result.validation_metrics
5	Rank the parameters by influence	sensitivity_analysis
6	Check each parameter, then the set as a whole	identifiability_analysis, analyze_subset
7	Write the values into the plant	digester.apply_calibration_parameters
8	Print the summary	-

Steps 5 and 6 belong together. Sensitivity says whether a parameter moves the output at all, identifiability says whether its effect can be told apart from that of another parameter. A parameter that fails either test should not have been in step 3, which is why check_identifiability=True moves the same question in front of the optimizer.

The record in step 1 is produced by simulating the plant itself with known parameter values, so step 4 can report the error against the truth rather than only "it converged". Its feed profile steps the organic load up and down and shifts the substrate mix, because a digester held at constant load settles into a state where several parameter sets produce the same gas curve.

The script does not persist anything. To keep a run, add Database.store_calibration after step 7, the way examples/scheduled_recalibration.py does it.