A sample size of one roast isn't enough to say anything useful, but it looks like the new code isn't worse than the old code, but is not as much of an improvement as I was hoping for. There's a parameter I can tweak before tomorrow's roasting to see if that helps, but I might need to investigate a different approach.