Would be great for data science courses if some of the code & data could be shared. I m currently designing one, and including these historical problems with tiny datasets would be a nice contrast to the current huge dataset/blackbox model approaches.

Thanks! I was thinking that there should be something more classical, like even in a parametric setting when your prior is off. Possibly in the context of Bernstein von Mises. That said I think you are right and you just go with the van der vaart book.

Berlin is annoyingly difficult to reach (+ you often have delays & problems) but going to south of France by tgv is unbeatable in my opinion. Fast, reliable & comfortable. Much better than flying.