Replay desk
The counts match. Are the records correct?
Keep the destination. Run the log twice.
Lumen · delivery log
Authored practice dataCount one impression per eventId in its occurrence hour. A viewer can have more than one impression. All times are UTC on one authored day.
| Delivery | Event | Viewer | Occurred | Arrived | Age |
|---|---|---|---|---|---|
| d1 | v1 | u1 | 09:01 | 09:02 | 1 min |
| d2 | v2 | u1 | 09:04 | 09:05 | 1 min |
| d3 | v1 | u1 | 09:01 | 09:06 | 5 min |
| d4 | v3 | u2 | 09:08 | 09:10 | 2 min |
| d5 | v4 | u3 | 09:59 | 10:00 | 1 min |
| d6 | v6 | u4 | 10:02 | 10:03 | 1 min |
| d7 | v5 | u5 | 09:12 | 10:45 | 93 min |
| d8 | v3 | u2 | 09:08 | 10:50 | 102 min |
The fast route accepts delivery age ≤60 minutes. Older deliveries take your late-data policy. This age rule is authored; it is not a streaming watermark.
What this model checks
The two jobs remove repeated keys within their own input. Append retains every emitted row; merge retains one row per eventId across both jobs and both runs. Counts follow occurrence hour. Matching this packet does not establish source completeness, production recovery or end-to-end exactly-once delivery. Your brief stays in this tab.
Netflix: recovery and duplicate boundaries