Published measurements for name-to-gender inference. Each report keeps coverage separate from accuracy, names the fixture, dates the run and shows weak segments instead of averaging them away.
A country and writing-system breakdown from one fixed 4,610-row fixture. The report includes zero-coverage segments and distinguishes declined answers from incorrect answers.
Definitions, sampling rules, commands and limitations. The separate holdout run uses names excluded from indexed evidence and reports its low coverage alongside answered accuracy.
Coverage uses all submitted rows. Answered accuracy uses only rows with a prediction. End-to-end accuracy returns to all submitted rows.
Country and writing-system results remain visible because a large Latin-script segment can conceal a serious gap elsewhere.
Data changes. Reports carry the actual generation date, and sitemap metadata changes only when the measurement does.
Unknown and wrong answers are part of the result. They are not removed to make the headline number more flattering.
The curated profile collection turns country-level evidence into citation-ready pages for names such as Andrea, Jean, Kim, Deniz and Ayşe. Each page exposes sample size and source beside the classification.
Explore name gender profiles →With your permission we measure how the site is used and which adverts lead to a sign-up. Both are optional and neither runs until you allow it. Read the cookie policy.
