VTR-1: The Verifiable Track Record Standard — Enforcing Backtest Honesty Without Trusting the Issuer or Revealing the Strategy

Abstract

Every correction for backtest overfitting depends on N, the number of strategies a researcher tried. The deflated Sharpe ratio needs it. So do multiple-testing haircuts and the probability of backtest overfitting. And N is supplied by the researcher. Nobody has ever been able to check it. VTR-1 is an open standard that makes N a structural property of a cryptographic proof. The full set of candidate strategies is committed before evaluation, and N is the leaf count of that commitment. The winning result is forced in-circuit to be the maximum of the committed set. The anti-overfitting statistics are then recomputed inside a zero-knowledge proof, on committed price data, net of a committed cost floor. Any party can re-derive every asserted property in milliseconds against their own copy of the data. No trust is placed in the issuer at any step, and the strategy is never revealed. The standard sets out three conformance levels, a verification procedure, and an adversary model naming seven attack classes. Its normative appendices pin the estimators, the fixed-point arithmetic and the commitment constructions closely enough that an independent implementation can reproduce a credential exactly. Forty-one numbered requirements let a conformance claim be contradicted. Nine known-answer conformance vectors are published with it, so a second implementation can be tested against something. The standard is equally explicit about what it will not certify: future returns, capacity, market impact, and any strategy whose result depends on fills the committed data cannot justify. The reference implementation's first published verdict refused its own author's flagship strategy, deflated below significance and published permanently. Published under CC BY 4.0. Anyone may implement it, including competitors of its steward.