Models & methods
Champion–challenger
Champion–challenger means one model is live (the champion) while others train on the same data and folds (challengers). If a challenger wins by enough, it takes the job. Loyalty is to the score.
This is how you avoid the annual model rewrite. The world changes in weeks. Bake-offs that happen in Q4 for a Q1 launch are already late.
Pinning is allowed. A human can freeze a champion for a regulated line. The tournament still runs in the dark, so you know what you are refusing.
Same folds, five candidates
Seasonal naive
16.4%
Smoothing
14.1%
Gradient boost
11.2%
Foundation
9.7%
Fine-tuned
10.1%