What this leaderboard cannot see

The coverage gaps and attribution uncertainty behind observed AI-assisted commits, and how to interpret the ranking responsibly.

The ranking describes observable public activity. It is not a complete account of AI use, a verified lower bound, or a measure of developer ability. Its main limitations affect both who appears and what a count can establish.

Private work is outside the measurement

Only public repository history is considered. Internal company work and other private development are absent, regardless of how much AI assistance they use.

Attribution practices vary

Tools and workflows expose different amounts of public attribution. Some assistance leaves no observable evidence. This makes comparisons sensitive to attribution conventions as well as actual activity.

Absence from a ranking does not establish absence of AI use. Differences between tool counts should not be interpreted as differences in adoption or market share.

Repository history is incomplete evidence

The available history may not preserve the full context of the work. A public commit cannot establish how much assistance occurred or which parts of a change were produced by a tool.

Scan coverage is incomplete

Repositories are scanned progressively. Work in repositories not yet scanned is missing from the current observation. Coverage reports the scope at the time of the snapshot; global coverage is not coverage for each person.

Attribution can be wrong

An observed association is not verified authorship. Incorrect attribution and identity linkage can inflate counts, while missed evidence can lower them. Neither the size nor the direction of error is guaranteed for an individual record. Suspected errors can be reported for correction.

Reading the ranking responsibly

Cite the metric, population, UTC window, snapshot date and scan coverage. Distinguish people from automation and the published subset from complete population totals. Interpret changes over time alongside changes in coverage and attribution practices.

Exact matching rules and detection examples are not published. The public methodology explains the measurement contract and its limitations without providing a guide to influencing detection.