# Verify a restore

Symptom: you restored from a snapshot or point-in-time and need to trust it.

## Diagnose what "done" means

Atlas marks the restore complete when the cluster is up, not when your data is verified. You need three checks: data completeness, index presence, and app behavior.

## Confirm

1. data URIs - compare document counts on key collections against the source (or the pre-incident counts). Spot-check recent documents around the restore timestamp; for point-in-time restores, confirm the cutoff is where you expected (oplog window limits how far back you can go).
2. Indexes: list indexes on the restored cluster and diff against the source. Restores should carry indexes, but verify rather than assume.
3. App: point a staging backend at the restored cluster and run the read smoke tests plus one write path.

## Fix

If counts mismatch, you restored the wrong snapshot or timestamp; redo from the right one. If indexes are missing, build them before cutover.

## Verify

Only cut traffic over after all three checks pass. Keep the old cluster until the new one has served cleanly through a full cycle, then decommission.