| name | data-loader-runbook |
| description | A safe data-load runbook — sandbox-first, backup, dedup, and verify — for any bulk data operation. Use before loading, updating, or deleting production data at volume. |
Data Loader Runbook
A disciplined, reversible procedure for bulk data operations so a load never becomes an irreversible production incident.
When to use
Any bulk insert/update/upsert/delete against an org — especially production.
Steps
- Sandbox-first. Run the full load in a sandbox that mirrors prod config; validate row counts and side effects (triggers, Flows, sharing recalc).
- Back up. Export the target records (and any cascade-affected children) before mutating. Confirm the export is complete and restorable.
- Map and dedup. Validate field mappings; dedup the source (and against existing records — prefer upsert on an external ID over blind insert).
- Disable noise deliberately. Consider deferring sharing calculation and, where safe and documented, pausing non-essential automation for the load window (
knowledge/large-data-volume-design.md).
- Load in batches. Use Bulk API 2.0 (
skills/bulk-rest-api-client) for volume; capture success/error files.
- Verify. Reconcile counts, spot-check records, confirm automation fired as intended, re-enable anything paused.
- Rollback plan. Keep the backup and the inserted-Id set ready to reverse.
Output
A filled-in runbook: environment, backup location, mapping, batch plan, verification queries, and the rollback step. This is high-blast — confirm prod actions with the human.
