Data safety
No AI suggestion can silently delete your original file
The planning model does not edit files. It proposes validated rules, and the product separates plan review from deterministic execution. The original upload remains a distinct object while each successful run writes a new output.
Destructive suggestions receive extra friction
Operations that can remove rows or columns have enforced risk floors. Even if a model labels one as low risk, server validation raises it to the required risk level. An enabled high-risk rule blocks plan confirmation until you explicitly acknowledge it.
You can skip the rule, edit its supported parameters, or leave the plan unconfirmed. There is no background mode that automatically accepts and runs a proposed cleanup.
The original and the cleaned result are separate
When you start a run, the Worker reads the source and writes a new output in the selected format. It does not save changes back into the uploaded CSV or workbook. A failed attempt records an error and leaves the source untouched.
The report makes removals visible through actual row, column, and rule-result metrics. Hashes and the immutable rule snapshot help identify exactly which source and plan produced a particular output.
- No execution during AI plan generation
- Confirmation required before a run
- Additional acknowledgement for high-risk rules
- Separate output and downloadable audit report
Cleaning a result is different from deleting a project
Cleaning rules can remove data from the new output, while project deletion removes stored project files and history after a separate destructive confirmation. The product does not confuse those actions.
Uploaded and output files remain available until you delete the project or your account. Preview-only lifecycle states are never presented as completed deletion events on real project pages.