Mathematical trust used to travel through text, reputation, and a familiar ritual: peer review.
But a growing class of “mathematics” is no longer just a written argument. It’s a toolchain: solvers emitting proof logs, proof assistants depending on massive libraries, numerical pipelines producing certificates, and environments that drift underneath everything.
In that world, trust becomes local. A result can be “true for us” because someone ran something once, on one setup, with one checker version. But it doesn’t travel.
Zero-trust is the upgrade: treat every prover — human, AI, solver — as an untrusted claim source until it ships verifier-facing artifacts.
That’s what I mean by Math Machines: not machines that find solutions, but machines that make solutions close.