This is a sequence of posts describing our team’s research into modularity: how to measure it, and which factors select for it in a variety of learned systems. The project began as part of the AI Safety Camp 2022 under the mentorship of John Wentworth, and has continued outside of this, with funding from CEA.