Proof at Scale: Math Machines for a World of Machine-Generated Conjectures
Mathine: Proof-Obligation Pipeline Machine
Link: https://doi.org/10.5281/zenodo.18706851
Machine-generated conjectures invert the historical bottleneck of mathematics. The limiting factor becomes less the scarcity of ideas and more the scarcity of auditable closure: proofs that can be checked, replayed, transported across contexts, and trusted under adversarial scrutiny.
This paper proposes Math Machines as the substrate for proof at scale—systems that treat verification as governed infrastructure, not a one-off scholarly event. The objective is not “more proofs,” but more trust that actually travels when proofs move across institutions, toolchains, and time.
The core model treats proof work as a pipeline of proof obligations governed by explicit admissibility rules, bounded verification budgets, and declared closure predicates. Instead of collapsing outcomes into “proved / not proved,” the system emits typed receipts that record what was assumed, what was checked, what was deferred, and why a result should—or should not—be treated as portable.
A key operational benefit is reducing proof debt. When obligations are made explicit and receipts are structured, partial progress becomes legible and upgradeable: you can see exactly what prevents closure, what must be rechecked after drift, and what can be safely reused. This lowers false-closure rates and makes the throughput of AGI/ASI-era conjecture generation compatible with legitimate verification.
The endgame is industrial, but not cynical: industrialize legitimate understanding. Each new summit should leave a trail others can audit, reuse, and contest without relying on reputation or narrative authority—so scale increases knowledge rather than just producing more uncheckable claims.
Mathine — Evidence In, Trust Out 