You're right. It's funny because I kind of noticed that, but with all of these subtle model issues, I'm so used to being distraught by the smallest thing I've had to learn to 'trust the data' aka the charts, model standings, performance, etc. and in this case, I was under the assumption 'it was the same model' clearly it's not.

Which is a bummer because it would be nice to try a true side-by-side analysis.