How Should Data Sources Be Attributed in an Export?
From the earlier reference collection. Read the current software blog.
Attribution should give a recipient enough information to identify the input dataset, its terms, and the process used to create the export.
Name the source precisely
Include the source's name and a stable location where recipients can identify it. A vague note such as “public data” does not tell anyone which publisher, dataset, or version contributed to the export. If several sources are combined, list each one and describe which fields or records it supplied when that distinction matters.
Record the source version, release identifier, or retrieval date when available. For a live endpoint without formal versions, note when the data was retrieved and, if relevant, the query or filters used. These details help explain why two exports made on different days may differ.
Include time and transformation
Distinguish the time the source data describes from the time your system retrieved it and the time the export was generated. These timestamps answer different questions. An observation from Monday can be downloaded Tuesday and exported Wednesday; a single timestamp cannot communicate all three events.
Summarize transformations that affect meaning, such as filtering, unit conversion, deduplication, aggregation, or coordinate changes. The summary need not reproduce implementation details, but should enable a reader to understand why the output may not match the source row for row. Preserve parameters or a transformation version when exact reproduction matters.
State reuse terms clearly
Include the applicable license name and a direct link to its authoritative text when known. Do not infer that data is free to reuse simply because it is publicly downloadable. If source terms differ across components, make that distinction clear and avoid implying that one license automatically covers everything in the export.
Keep attribution visible near the download and inside the exported artifact where practical. A README, metadata file, or dedicated attribution fields can travel with a bundle. If a single CSV cannot carry rich metadata, provide a companion file and make the relationship obvious in the download instructions.
Make attribution travel
An export often gets copied away from its original page. Include enough context in the file bundle for someone to identify the publisher, source dataset, retrieval time, license, version, and transformation. Avoid relying on a web page that may not accompany a downloaded file.
Before publishing, verify that every cited source actually contributed to the output and that the license link points to the relevant terms. Good attribution supports responsible reuse and helps users investigate differences without treating the export as an unexplained snapshot.