Seven column-level checks before you export a clean dataset
The review checklist we use to catch the most expensive normalization mistakes before they leave the browser.
Most broken exports are not caused by dramatic failures. They come from one unchecked assumption in one column that looked harmless during a spot check.
A lightweight pre-export review keeps that risk manageable and makes the resulting file much easier to trust.
Run through the same column-level checks regardless of file type. The goal is consistency, not cleverness.
- Type: confirm the inferred type matches the real business meaning.
- Format: verify date, number, currency, and percentage conventions.
- Nulls: inspect blanks, placeholders, and sentinel values such as NA or dash.
- Ranges: catch outliers, impossible values, and negative numbers where they should not exist.
- Categories: check casing, spelling drift, and duplicate labels that should collapse to one value.
- Identifiers: preserve leading zeros and do not coerce keys into numbers.
- Output shape: confirm the exported representation matches what downstream tools expect.
A strong review step does more than find malformed rows. It surfaces uncertainty. If a column could reasonably be interpreted in two ways, that is exactly when you pause and set the rule explicitly.
The cleanest export is the one that leaves the fewest decisions for the next person or service in the chain.
Apply It
Review the column rules before you transform the entire file.
Normalize is built around that workflow: inspect a sample, confirm the meaning of each column, then export a clean dataset with explicit output settings.