Bolzano paper claims about 200 open problems solved without problem-specific guidance

Applying the system to about 3,800 open problems without tailored guidance

The paper describes Bolzano as a system using parallel prover agents, a verifier agent, and a human-readable research state. Its authors say they applied it to about 3,800 open problems drawn from four sets of papers. In an experiment involving papers accepted to STOC 2026, they say Bolzano answered four questions raised in the papers, with confirmation from the papers’ authors.

The paper describes an initial manual use of Bolzano on expert-selected problems, producing eight results whose proofs were checked by domain experts.

The authors report running Bolzano without problem-specific human guidance across about 3,800 open problems and solving about 200 of them.

The reported work extends from expert-selected, manually guided proof search to a broader run on open problems without problem-specific guidance. The stated results remain unverified outside the paper.