Assuming you are talking about the Castle Bravo test, it was 2.7 times larger than predicted not 1000. I can't think of any test offhand that was 1000x off the predicted yield, unless you count the fizzles that might have been 1/1000th predicted yield. Castle Bravo was a terrible disaster, but more from the fallout than the actual explosion. The models of fallout and its propagation were much more inaccurate than the problems with the yield calculations, and the source of the tragic consequences.