McNemar's paired test for comparing two NLP models evaluated on the same test set.

Live: https://stresearch-dev.github.io/mcnemar-nlp/

Automatically switches to the exact binomial (sign test) when b+c < 25, following Dietterich (1998). Includes Edwards' continuity correction toggle, one- and two-tailed modes, and paper-ready APA and LaTeX output sentences.

Single HTML file. No dependencies. Runs entirely in the browser. MIT licensed.

Reference check: b = 22, c = 2 → exact binomial p = 3.59e−5.

Part of the LLM Evaluation Toolkit by Shubham Tibrewal.