Data provenance

Data provenance is the documented history of a dataset, where it originated, what instrument, survey, or process produced it, and every transformation, filter, or correction applied to it since, up to its current state.

Provenance can be captured manually, through documentation and version control, or automatically, where a scientific workflow or data pipeline records each processing step as it runs. The term overlaps heavily with data lineage in a business data context, but provenance is the term more common in scientific and archival settings, and it typically places more weight on capturing the scientific method and instrument conditions behind the data, not just its technical processing path.

A result is only as trustworthy as the chain that produced it, provenance lets another researcher, a journal, or a regulator verify how a number was actually derived, which is central to reproducibility and to catching an error that crept into a widely reused dataset before it spreads further. It is also foundational to practical research data management. Pitfalls include provenance recorded after the fact from memory, which is unreliable, and datasets copied, subsetted, or merged across sources without retaining that history, which effectively loses its provenance even if each individual step along the way was sound.