Preregistered Backtest

A preregistered backtest is a validation whose population, cutoff date, metrics and pass criteria are written down and locked before any result is generated. It prevents the test from being reshaped after an unfavorable early result.

The registration records what will be tested, on which matters, against which cutoff, using which metrics and what result counts as a pass or a failure. Once filed, it cannot be edited. The backtest is then run once, and the result is reported against the registered criteria whatever it shows.

Preregistration is what makes a claim per asset class defensible. A claim that a score is valid for an asset class activates only when a model trained on the relevant realized outcomes passes its registered backtest, and each activation is published with its evidence.

A preregistered backtest is not a guarantee of future accuracy. It establishes that a method met a stated standard on matters it had not seen, and it leaves a record that a reviewer can check.