But of course there's a huge RL literature too, and I'm eager to figure out the boundary conditions for RL vs. habits.

in addition to Miller et al 2019, there's been Greenstreet et al 2025 (Nature), Wagner et al (2024 preprint), and in some interpretation Eckstein et al (2024 preprint), ending up on very similar theories.

Important question, and I'm not sure yet. We're currently trying to show evidence of both at the same time, and it's a challenge.
It's been interesting seeing the habit idea coming out in many recent papers from different angles.

Yes, I'm interested in figuring out the discrepancy. I do have unpublished test phase data where the WM+habit does capture the test phase better than WM+RL. Agree it's important to figure out what's different, and under what circumstances RL is needed. Working on it!

Thanks Chris. I haven't gone back to the data you're referring to, but I have gone back to the individual differences data (clinical and developmental), and most conclusions are unchanged - happy to follow up by email if you're interested!

@jennifersenta.bsky.social's paper is now published!
Check it out here!
journals.plos.org/ploscompbiol...