Glossary
Pre-registration
Writing down a research plan, with its pass mark, before looking at the results. It separates a test of an idea from a search for one.
Pre-registration is a dated record of the question, data, definitions and pass mark, made before the analysis is run. It limits the freedom to choose a pass mark or a variant after seeing what worked, and it separates predicted results from findings made afterwards.
It does not prove an idea is right. It limits bias.
Example: a plan states that a rule must show a t-statistic above 3.0 on a holdout it has not seen. If the holdout gives 1.8, the idea is rejected, and the threshold is not lowered to 1.5 afterwards.