Issues with parameter recovery

I guess my first suggestion would be to “unbuild” the hierarchy and try to run the model with a single subject-worth of data and see where things stand at that point. I find debugging more complex models difficult and often find that simplifying helps. Have you tried that?