Scheduler trigger

Load current watermark

System A: fetch since watermark

Retry - exp backoff on failure

Capture A: data + success flag

System B: fetch since watermark

Retry - exp backoff on failure

Capture B: data + success flag

System C: fetch since watermark

Retry - exp backoff on failure

Capture C: data + success flag

Aggregate successful fetches

Mark partial fetch; log + metrics

Proceed with available data

Group by recordid

On Error Propagate: alert and stop run

Select latest record per group by timestamp

For each latest record: fan out to target systems

Retry with backoff

On Error Continue: send to DLQ - recordid, system, payload, error

Mark update success

Collect update results

Advance watermark to max processed timestamp

Skip watermark update; will re-fetch window

Emit metrics and finish