OptionalDefaultOptional default dry-run sample cap for this processor when none is explicitly configured.
OptionalFailWhen true, a run that completed with one or more failed records ends with status Failed.
OptionalBuildOptional hook to build a processor-specific run summary (and dry-run refusal reason) at the end of a run.
OptionalComputeOptional hook to compute a custom basis hash for watermark change detection.
OptionalPreflightOptional run-level readiness check, called ONCE before the first batch. Throw to refuse the
whole run with a clear, actionable message (e.g. "this model needs training") instead of
failing every record one by one until the error-rate circuit breaker trips. The run is still
recorded (status Failed, the thrown message as its ErrorMessage) so history stays honest,
but no record is processed. Processors without a run-level prerequisite simply omit it.
The run-level execution context (user, provider, run/process IDs).
OptionalPrepareOptional hook to enrich output mapping configuration and report whether the processor has unmapped writes.
OptionalProcessOptional batch execution seam (P1-7c). Enables two-phase distinct-key execution: deduplicating across the batch before execution, executing once per distinct key, and fanning results back to all matching records.
The batch of records to process.
The execution context.
Either:
Map<string, RecordResult> keyed by RecordID.RecordResult[]. When returning an array, it MUST be positionally aligned 1:1
with the input records array (same length and order).Processes one record.
The record to process.
The per-record execution context.
The record's outcome (Succeeded / Failed / Skipped) plus any payload / trace links.
PROCESSOR (executor) seam — does the work for a single record and returns its outcome. Concrete processors wrap an Action, an Agent (via Execute Agent), an Infer-&-Write-Back step, or a plain function.