Compared · as we understand their public materials · corrections welcome

MIZAN and vBase: what each one proves.

We name competitors and neighbours, because hiding them is what weak positions do. This page states what vBase proves, what MIZAN proves, and where the two sit relative to each other.

What vBase proves

vBase proves when a dataset or result existed and that it has not changed since: a hash plus a timestamp. That is commitment, and it is valuable.

What MIZAN proves

MIZAN proves the computation itself was honest: returns after costs, the out-of-sample split, and the overfitting corrections recomputed inside a zero-knowledge proof, with the trial count bound as a Merkle leaf count. A commitment says a claim was made at a time; a credential says the claim survives examination.

Where they sit

Complementary, not competing. A record can be committed with a timestamp service and judged by MIZAN; MIZAN itself anchors every credential to Bitcoin and RFC-3161 before judgment.

What nobody on this list does

To our knowledge, no one else proves the anti-overfitting framework inside the proof itself: the deflated Sharpe ratio with the trial count as a committed Merkle leaf count, recomputed in-circuit. That is the unclaimed square, and it is the one the standard occupies. If that changes, this page will say so.

Other comparisons: Proven · Knidos · AuditZK · Darwinex · Numerai · QuantConnect.