How it works

Profiling and data quality

What the profiler checks, and why data-quality problems are findings of their own.

Profiling types every column (dates, currency, percentages, categories, IDs), normalizes values, and looks for problems a careful analyst would flag before trusting any number:

Unit and scale errors — a value 1,000× its neighbours (the coffee example’s February packaging cost).
Ambiguous dates — 03/04/2026 read both ways; resolved from rows where the day is above 12, and reported either way.
Duplicates, gaps and future dates — repeated keys, missing periods, dates after “today”.
Entities — “OOO Hotel Group” and “Hotel Group LLC” linked as one customer.
Injection screening — table labels and text cells are screened for instructions addressed to a model before any model sees a sample.

Each problem becomes a data_quality item in the report, with the cells or rows it concerns. Findings never silently “fix” the data.