The answers returned by the decision runner.
The candidate notes evaluated.
The model that answered: the MJ decision model and the model its driver resolved.
The calibrated probability at which a note is kept (default: MEMORY_NOTE_MIN_PROBABILITY, 0.6).
Optional calibration table override (defaults to MEMORY_NOTE_DECISION_CALIBRATION).
Evaluates decision answers against candidate notes using Platt calibration.
The calibration is the one fitted on the exact model that answered (
FindDecisionCalibration). An answer from any other model, including a calibrated model at another version, is uncalibrated and keeps nothing (fails safe toward writing no durable memory).