Early days for this project but the idea is interesting. I could clearly things I'd want to automate that way.

We already had reproducibility issues in science. With such models which allow to produce hundreds of "novel" results in one paper, how can we properly keep up in checking all the produced data is correct? This is a real challenge.