Glossary
Meta-labelling
A second model that decides whether to act on the signal of a primary rule, by predicting if that signal will be right.
In meta-labelling a primary rule picks the side of a trade (long or short). A secondary model then predicts whether that particular signal will succeed, and the output is used to skip signals or scale their size. The term comes from López de Prado (2018).
It filters a rule that already works; it does not create one. It reduces the number of trades, so it must raise expectancy enough to matter after costs.
Example: a filter that takes 150 of 400 signals and is right 60% of the time in testing changes the win rate among taken signals from 44% to 60%.