This pull request introduces the new automated calibration workflow for ActivitySim, modularizing calibration logic into a dedicated package and integrating it into the main simulation runner. The changes include the addition of a comprehensive calibration package, updates to the CLI to support calibration runs and output preservation, and enhancements to model settings validation to include calibration configuration.

Calibration package introduction and integration:

• Added a new activitysim.core.calibration package, splitting calibration logic into focused modules such as orchestrator.py, settings.py, component.py, and more, with a clear public API. [1] [2] [3]
• Updated activitysim/cli/run.py to import and utilize the calibration package, including logic to determine when to preserve outputs and to run the calibration workflow before standard or multiprocess simulation. [1] [2] [3]

Settings and validation improvements:

• Enhanced model settings validation in settings_checker.py to explicitly validate the optional calibration.yaml configuration file using the same error aggregation as other settings. [1] [2]

Random number generator and reproducibility:

• In initialize_households, ensured that both households and persons dataframes are registered as random number generator channels to support reproducible calibration and simulation.

Documentation:

• Added a detailed README.md for the calibration package, describing its architecture, configuration, iteration/recovery semantics, and integration with the main workflow.