That this feature works as well as it does should cast severe doubt on any talk about the shape of a graph of rate of change over time. It's quite common that if you calibrate multiple roasters using this technique you'll have runs where the raw data and various calibrations for the same roast have completely different shapes. This work predated anybody talking about that and the best later research I'm aware of only seems to strengthen my argument.

https://video.typica.us/videos/watch/dee5ba1a-f455-45af-b8bb-cb17a6650d30