**TL;DR:** Restore to a new cluster, verify data and app behavior, then cut over. Never restore over production. It must be inside the continuous backup oplog window; outside it, Atlas cannot do it. Restore to a NEW cluster, never over the production cluster.

## The fix

**TL;DR:** Restore to a new cluster, verify data and app behavior, then cut over. Never restore over production. It must be inside the continuous backup oplog window; outside it, Atlas cannot do it. Restore to a NEW cluster, never over the production cluster.

## The fix

1. Pick the restore point: the exact timestamp before the bad event. It must be inside the continuous backup oplog window; outside it, Atlas cannot do it.

2. Atlas UI: Backup, cluster, Restore, point in time. Restore to a NEW cluster, never over the production cluster. Restoring over production destroys the evidence and the rollback path.

3. Wait for the restore to finish. Large clusters take a while; the oplog replay is the long tail.

4. Verify: document counts on key collections, spot-check documents around the cutoff timestamp, confirm indexes exist, run the app smoke tests against the new cluster.

5. Cut over: update the connection string (env var, rolling deploy), watch metrics on the new cluster.

6. Keep the old cluster for a retention period before deleting, in case the restore point was wrong.

7. Decide the restore point from evidence (logs, timestamps), not gut feel. Off by an hour can mean restoring the corruption too.

8. Tell the team before cutover: writes to the old cluster after the restore point are lost. Freeze writes during the window.

## When to use this

- This covers exactly what the title says: Atlas.
- You are setting this up for the first time, or auditing an existing setup.
- You want the key gotchas in one place before you start.

## When not to use this

- You are doing a different workflow with Atlas; these steps are specific to the title above.
- You need the full reference docs; this is the short path, not the manual.

## Compatibility

- Not pinned to a specific version; follows current Atlas behavior.