A quick crossref to the ever knowledgeable @junpenglao would suggest that yes: it’s common practice to loop the LKJ priors LKJCholeskyCov input dependent
I hear you on the not quite elegant bit, and to jump on from Junpeng’s note, with your participants x trials dataset, I hazard that you might be able to treat this as a pm.MatrixNormal, though TBH I’m not familiar with them in practice. I found a good explanation here: Matrix Variate Normal Distributions with MixMatrix
Re uij, if you’re setting a different chol per participant (is this i ?), then wouldn’t;t you have a uj? FWIW you can probably index 2D if you want to u[i][j], but that’ll probably be a level of complication you dont need