Math AI LabTheoremSearch/Help us Improve

Help us Improve

TheoremSearch is an ongoing research project. Your feedback directly shapes how the search model is evaluated and improved.

Relevance feedback

Every search result has a pair of feedback buttons in the top-right corner of its card:

Thumbs up

This result is relevant. The theorem matches what you were looking for.

Thumbs down

This result is not relevant. The theorem does not match your query.

Each vote is recorded once per result. We store the query, the theorem, and your vote. We do not store personal information. See our Privacy Policy for details.

Reporting issues

Each result also has a flag button. Use it to report problems with individual theorem entries:

Report an issue

Opens a panel with selectable reasons: the slogan and precise statement don’t align, LaTeX is malformed, the slogan is not descriptive, the theorem type is incorrect, or other.

Reports are used to find and fix systematic issues in slogan generation and theorem parsing. We only store the selected reasons and, if provided, additional description.

How it helps

Votes give us ground-truth relevance labels that we use to measure and benchmark retrieval quality. This will guide future improvements to the embedding model, ranking, and slogan generation pipeline.

Even a small number of votes on diverse queries is valuable. If you use TheoremSearch regularly, taking a moment to vote on results is one of the most useful things you can do.