To the victims I've previously hassled to provide individualized analyses in papers, I practice what I preach: Fig S1 is a 64-page figure with visualizations of LangFC (blue) vs task (red, magenta overlap) in every one of our nearly 2,000 sessions. This one is me 👇!

...And reproduces localizer task-based results

...Can be identified efficiently

...Is stable within individuals and variable between them

...Is robust to the granularity of the parcellation

...Differs in function from nearby networks

...Emerges across task states

...Is selective for language

...Shows expected topography, distinct from other brain networks

Task-agnostic iFC could let us skirt both of these issues, and when we used it, we systematically found that one of the networks in the parcellation (LangFC, identified fully unsupervised) turned out to be the language network (Fedorenko et al 24). In particular, LangFC...

P.S. - There's currently a bug in which Fig 2B is missing its legend. The editorial team is working on it, but meanwhile, here's how the fig is supposed to look:

6/

We show in SI Appendix 1 that their linear-in-predictability result depends on the use of cloze predictability estimates. Model-based predictability yields a logarithmic pattern.

5/10