the model + target + scope (+ optional write-back column); see CreateScoringProcessOptions
the saved Record Process AND — in write-back mode — its scoring binding
(null in generic mode); see CreateScoringProcessResult
if a required input is missing, the scope is not exactly one selector, the target entity is unknown, the Record Process fails to save, or — in write-back mode, after the Record Process was saved — the binding fails to save (the message carries the cause). The binding save is intentionally fail-loud (not swallowed): a saved RP without a binding would run invisibly to the lineage UX, so we surface the inconsistency.
Create and save an on-demand Record Process that scores
targetEntity's rows withmodelId— no schedule. The generic "Run Record Process" run-now Remote Op and the generic scheduling dialog target the returned row later. Operates in one of two modes, driven by whetheropts.outputFieldis supplied:outputFieldpresent): a run writes the prediction into that target column (via the Record ProcessOutputMapping), AND this helper records aMJ: ML Model Scoring Bindingslineage row (Mode='OnDemand') tying the model → Record Process → target entity/column so the operationalized model is discoverable by the UX surfaces that read those bindings.result.bindingis the saved row.outputFieldomitted): noOutputMappingis set and no binding is created — a run's predictions are recorded in the process run history (MJ: Process Run Details) only.result.bindingisnull.Unlike the scheduled sibling, this row carries
ScheduleEnabled=falseand NO cron, so saving it does NOT create an ownedMJ: Scheduled Jobsrow — it stays purely on-demand until a generic scheduler dialog (optionally) schedules it.