Audit before you model, always
Almost every surprising result we are asked to review turns out to be a data problem: duplicated submissions, a column that silently became text, or missing values coded as 999.
The checker runs entirely in your browser, so you can audit sensitive data without uploading it anywhere.
Read missingness by column, not overall
An overall 4% missing rate can hide one variable that is 60% empty. Sort your attention by the per-column missing percentage and decide, per variable, whether to impute, drop the variable, or drop the cases.
State your rule in the methodology before you apply it.
Treat duplicates as a design question
Duplicate rows can be genuine repeated measurements or the same respondent submitting twice. The count tells you there is something to inspect; your questionnaire design tells you which it is.
Investigate outliers, do not delete them
The report flags values beyond 1.5 times the interquartile range. That is a screening rule, not a verdict. Check whether the value is an entry error, a legitimate extreme case, or a unit mismatch.
Deleting genuine extremes to make a model behave is how findings become unreproducible.
Confirm the detected types
If a numeric variable is reported as categorical, something in the column is not a number — a stray comma, a currency symbol, or a text note in a numeric field. Fix it at the source file, then re-run the audit.