Prompt library
Plan a data cleaning pass before touching the data
Profile-before-fix stops the classic error of cleaning problems the data does not have while missing the ones it does. Reversibility labels and the quarantine rule preserve the audit trail that separates defensible analysis from data laundering.
Last reviewed July 17, 2026
The prompt
Plan the cleaning of the dataset below before any transformation happens.
Dataset description (columns, types, source, rough size): {{dataset}}
What the clean data must support: {{purpose}}
Produce:
1. Audit checklist first: per column, what to profile (null rate, cardinality, min/max sanity, duplicate keys, encoding traps, unit consistency). Output as a runnable checklist.
2. Predicted dirt: given the source, the 5 most likely quality problems and the query or check that would confirm each.
3. Decision table for each anticipated problem: fix rule, the records it touches, and whether the fix is reversible. Destructive fixes (dropping rows, collapsing categories) require a stated justification tied to the purpose.
4. The quarantine rule: what happens to records that fail validation (never silent deletion; a flagged quarantine table or column).
5. The log: what the cleaning script must record so the numbers in the final analysis can be traced back through every fix.
Rule: no fix without a prior profile confirming the problem exists at meaningful scale.Run in idaptOpens a new chat with the prompt prefilled. Nothing sends until you press send.
Fill in the variables
| Variable | What it is | Example |
|---|---|---|
| {{dataset}} | Columns, source, size | CRM export: 90k rows, company, contact, deal_size, stage, created_date; hand-entered by 14 reps over 3 years |
| {{purpose}} | What the clean data feeds | win-rate analysis by segment for the sales review |