Experimentation

Experimentation is how you test a change to a recommender or an experience before rolling it out to everyone — you split traffic across variations, measure a chosen metric, and let the data decide the winner rather than guessing.

In this section

Create and manage experiments

Set up an experiment to test personalization variations and measure which performs best.

Compare recommenders (worked example)

Run an experiment that pits two recommenders against each other and read the result.

Discussion

Comments are not configured for local development. They appear once the site owner connects GitHub Discussions (Giscus).