Representative sample
By Gareth HoyleLast updated How we research
A representative sample is a subset of a dataset drawn so its mix of record types, dates and quality reflects the whole archive, not hand-picked highlights.
A representative sample is a subset of records selected so that it reflects the full dataset in content, date range, quality and format. It lets a buyer judge the whole archive without seeing all of it.
How should a representative sample be drawn?
A representative sample should be drawn at random across the archive, not hand-picked. Stratified sampling, where records are drawn in proportion from each system, year or record type, makes sure small but valuable segments appear. Label the sample with how it was drawn.
Why do buyers distrust samples?
Buyers distrust samples because a seller can show the best records and deliver the rest. Contributors to a 2026 Hugging Face forum thread on selling datasets described this as a trust deadlock and suggested labeled stratified samples, paid pilots, acceptance tests and staged payments to break it.
A representative sample is one of the eight points in our seller vetting standard. See how to evaluate a dataset sample and hold-out set.