M5 · Did It Actually Work?
Every Example Gets a Turn
Run k-fold cross-validation and report the mean with its spread instead of one number.
The same model on the same data scored 80 and scored 95, in the same minute. Nothing changed except which fifth of the posts was held back.