the injected runtime seams the ML scorer needs (model loader, artifact loader, sidecar)
OptionalautoTrainer: IModelAutoTraineroptional auto-trainer for train-if-untrained processes
OptionalshadowCoordinator: ShadowScoringCoordinatoroptional shadow scoring coordinator for challenger runs; defaults to a production coordinator wired with MLModelChallengerScoreRunner and FeatureValueShadowAuditSink
Register the MLModelInferenceProcessor into the Record Set Processing RecordProcessorRegistry for the
'ML Model'work type (and the'MLModelInference'alias), soRecordProcessExecutor.buildProcessor()resolves it automatically — WITHOUT the substrate depending on Predictive Studio. The registered factory closes over the injected MLInferenceDeps (model loader / artifact loader / sidecar) and reads the per-run MLScoringConfiguration (themodelIdetc.) off the Record Process'sConfigurationandChallengerConfigurationObjectoffcontext.RecordProcess.Idempotent — calling more than once simply re-registers (last-wins). Call from the engine's startup/bootstrap with the production deps.