Update: a physicist's BootLoops run—36 papers across 18 fields
The open-source provable-math harness gets its first big result—one physicist, Claude, 3 months, and 50+ hours of human checking per project.
Copy markdown
36 manuscripts, 18 fields, one open-source loop
Harvard quantum-field theorist Matthew Schwartz reportedly used BootLoops with Claude to draft 36 manuscripts across 18 fields—from particle physics to linguistics—in about three months. BootLoops forces the model to run exact, re-runnable calculations instead of plausible-looking math, so the output is checkable rather than vibes.
The catch: 50+ hours of human review per result
Schwartz's rule was “look at everything yourself”—he logged 50-60 hours supervising a single physics paper. BootLoops is open source, so you can bolt the same provable-calculation loop onto your own agent this weekend; just budget real review time, because the unlock is reproducible math, not hands-off science.