All setups
DataSetupAI-generated

Data Cleaning Plan

Turn a messy dataset into a step-by-step cleaning plan with verification checks

What it is

Turn a messy dataset into a step-by-step cleaning plan with verification checks

Produce a structured data cleaning plan from a description of the dataset and its quality issues. Triages problems by severity, specifies detection method and fix logic for each, and ends with a row-level audit to confirm the clean dataset is correct.

What you'll get

Custom instructionsCompiled from your answers, editable before you export.
Data cleaning checklist and reference cardGives Claude a structured triage framework and a standard set of checks to run on any dataset, so cleaning plans are thorough and consistent rather than addressing only the issues the user already noticed.
Export bundle for Claude ProjectsCopy-paste blocks with a guided walkthrough.

Try it on

Clean a CRM export with date and country issuesscenario · built in
Claude should triage the three issues: missing email (major — affects any email-based join), inconsistent dates (blocking — analysis on date columns will fail), country naming (minor to major depending on downstream use). For each it should specify detection (e.g., regex to identify format), fix logic in the chosen cleaning tool, and a verification step. It should ask whether the 800 missing email rows should be excluded or retained with a null flag.
Clean an events table for retention analysisscenario · built in
Claude should triage: duplicate event IDs (blocking — retention calculation will double-count), orphaned user_ids (major — retention denominators will be wrong), local time timestamps (blocking — cohort assignment will be wrong across timezones). It should specify a deduplication strategy (keep the first event within the 1-second window), a referential integrity check, and a timezone conversion approach. The plan must end with a row-count audit.

Compiled preview

You are a data cleaning plan writer. Produce a structured, actionable cleaning plan. Data source: {{dataSource}} Downstream use: {{downstreamUse}} Cleaning tool: {{cleaningTool}} Known quality issues described by the user: {{knownIssues}} Rules: 1. Triage every issue by severity: blocking (analysis is wrong without fixing), major (results are biased), minor (cosmetic). 2. For each issue specify: detection method, fix logic in {{cleaningTool}}, and how to verify the fix worked. 3. End the plan with a row-level audit — expected row count, primary key uniqueness check, and value range checks for key columns. 4. Never recommend silently dropping rows without flagging the volume lost and its p