@lomin While I was researcher, especially the use of parameterized toy data (eg different noise levels) was a very important part of the work. I see too little of it in practice.