As a scientist who works on computational models, this kind of thing is terrifying.

There's just so many ways that writing software can go awry that I have a hard time believing any computational result too strongly.

I think you could give this a subtitle of "Why you should read the Art of Computer Programming, Vol 2." Number representation in computers is a vast field that really deserves all of the attention it can get.

Note that this is fairly common in rodent models for human diseases [1], but as the referenced article says: "Less than Perfect, Still Invaluable". I guess the same applies here.

[1] http://www.hopkinsmedicine.org/institute_basic_biomedical_sc...

I gotta say, I love to be working in San Diego these days. The deluge of results coming from here or originating here and being re-examined elsewhere (as in this case) is just great.