Personally, I think it’s great that people dig out old threads to add information, that way other people can find everything in one place. But not quite sure what the actual policy is there…
You are certainly not the only one that fell into this trap. What did happen in your case? Insane runtime scalings when adding more intervals as e.g. here? And/Or runtime was fine, but the chains were stationary?
I do wonder whether it’s mainly the bad runtime scaling or also the bad gradient approximation. I’ll link back and forth to `log_prob_grad` method for CmdStan mimicking its `generate_quantities` method