toolEnglish

Governance and Best Practices

How to profile responsibly, protect sensitive data, avoid common mistakes, and apply a minimum review checklist.

EnglishEspaƱolFranƧais

Documentation follows the current system language. Missing translations fall back to the page's canonical language.

Governance And Best Practices

Data Profiling is most valuable when the team treats it as a governed review, not a casual upload.

Responsible profiling loop

Before Profiling

  • Confirm that the workspace is the right place for the dataset.
  • Remove fields that are not needed for the decision.
  • Check whether personal, confidential, or regulated information is present.
  • Confirm file size and row limits.
  • Add descriptor notes so reviewers understand definitions.
  • Record source, owner, period, and filters.

During Review

  • Start with missingness, duplicates, and obvious type problems.
  • Inspect columns with quality flags before discussing findings.
  • Compare chart suggestions with the decision question.
  • Do not treat inferred type as authoritative if the domain meaning differs.
  • Keep notes about assumptions and unresolved issues.

After Profiling

  • Save the draft or create a new version before using the profile in another decision.
  • Document whether the dataset is ready, requires cleaning, or is not fit for the question.
  • Link the profile to the decision model or report that uses it.
  • Lock the model when the profile becomes evidence for a formal review.

Privacy And Sensitivity

Profiling can reveal sensitive values through top-value lists, identifiers, and outliers. Avoid uploading direct identifiers unless needed. Consider masking, aggregating, or removing fields before profiling.

Common Mistakes

  • profiling a file without confirming its period or filters
  • assuming missing values are random
  • deleting outliers without domain review
  • using identifier columns as explanatory variables
  • presenting exploratory charts as causal findings
  • failing to save a version before cleaning or replacing the dataset

Minimum Review Checklist

Before the profile is used in a decision, confirm:

  • source and period are documented
  • missingness is acceptable or explained
  • duplicate rows are understood
  • key columns have definitions
  • high-severity signals are resolved or documented
  • recommended cleaning actions have owners
  • the profile is saved as a version