OptionalActionDeep-trace link to an Action Execution Log, when the work was an Action.
OptionalActionResolves once the MJ: Action Execution Logs row named by ActionExecutionLogID has been
written. The Action engine writes that row fire-and-forget, so the ID can name a row that does not
exist yet. A tracker that stores ActionExecutionLogID in a foreign key must not save the row
before this resolves; it should queue the wait with the save rather than await it here, so the
per-record loop never blocks on the log write. RecordSetProcessor.Process also awaits it once,
before CompleteRun, so a run never returns while a log row is still being written, whatever the
tracker. Never rejects. A live promise, never serialized; set only by the Action record processor.
OptionalAIDeep-trace link to an AI Agent Run, when the work was an Agent.
OptionalAIDeep-trace link to an AI Prompt Run, when the work was an Infer-and-Write-Back.
OptionalAttemptNumber of attempts made for this record.
OptionalBelowThe outputs answered below their confidence floor and withheld.
OptionalConfidencePer-output confidence scores keyed by output Name: the driver's own, else the answer's at each ConfidenceRef.
OptionalDryTrue when a dry run computed the record (nothing written, no Outcome).
OptionalDurationProcessing duration in milliseconds (the engine fills this in if the processor does not).
OptionalErrorError detail when Status is Failed.
OptionalFailureSet when the answer was rejected (see RecordFailureKind): the run counts it apart from other failures.
OptionalFeatureReference to the FeatureValueCache entry when served from or saved to the cache.
OptionalHumanOptionalOutcomeWhether this record's values and write-back changed anything. Absent on a dry run.
OptionalPipelineFeature Pipelines: the pin the record was computed under; the run result reports it.
OptionalPipelineOptionalPromptThe MJ: AI Prompts row behind AIPromptRunID and PromptVersionHash, when it is not the Record
Process's own prompt (an escalated record is answered by its target prompt).
Absent means the process's own prompt.
OptionalPromptContent hash of the prompt version / instructions used to compute this record's result.
OptionalResultStructured output payload produced for the record (persisted as JSON on the detail row).
Whether the record succeeded, failed, or was skipped.
OptionalWarningsNon-fatal problems with this record's processing (e.g. its cache entry could not be stored).
Per-record outcome returned by
InferProcessor, extending RecordResult with Feature Pipeline pin metadata.