Recently I reviewed a paper in which most proofs were done in a proof assistant. Yes, the machine guaranteed that the proofs were correct, but I still had to make sure that the authors correctly formulated their definitions and theorems, that the code did not contain hidden assumptions, that there were no unfinished proofs, and so on.

In a typical situation an author submits a paper accompanied with some source code which contains the formalized parts of the work. Sometimes the code is enclosed with the paper, and sometimes it is available for download somewhere. *It is easy to ignore the code! *The journal finds it difficult to archive the code, the editor naturally focuses on the paper itself, the reviewer trusts the authors and the proof assistant, and the authors are tempted not to mention dirty little secrets about their code. If the proponents of formalized mathematics want to avert a disaster that could destroy their efforts in a single blow, they must adopt a set of rules that will ensure high standards. There is much more to trusting a piece of formalized mathematics than just running it through a proof checker.

Continue reading How to review formalized mathematics →