Run started.
Run terminated successfully.
Run terminated with failure.
Run cancelled mid-flight by a user/system abort (not a failure, not a clean completion).
Run resumed from a prior checkpoint after a kill/restart.
A named stage within the run started.
A named stage within the run completed.
A named stage within the run errored (run may continue or fail).
A batch of records is about to be processed (sync paths).
A batch of records has been processed.
A single record errored within a batch.
A non-fatal warning surfaced during the run. Carries a structured
{stage, code, message, data} payload (see SyncWarning). Distinct
from stage.error/record.error: a warning never fails the run, it is
aggregated separately into the result's warnings[] rollup.
Long-running progress with no record-level granularity.
Resumable checkpoint. Carries resumableState — the subsystem-specific
shape needed for the run to resume from this exact point. On restart,
the resumption engine reads the latest checkpoint event for any in-flight
run and hands the resumableState back to the originating service.
An outbound API call started (vendor-API debugging, esp. GraphQL).
An outbound API call completed.
An outbound API call retrying after transient failure.
TransformRecord (or equivalent) reshaped a record.
Discovery added a new IntegrationObject row.
Discovery added a new IntegrationObjectField row.
PK classifier invoked for an object.
PK classifier returned a verdict.
An MJ entity was generated for an IO row that has a PK.
An IO row was skipped from entity generation because it lacks a PK.
Event types in the progress.jsonl stream.