Designing scientific components

Correctness is more precious to scientific programmers than it is to business programmers because of the great difficulty in distinguishing between programming errors, errors in modeling, and errors in algorithms. We've all sat in meetings and discussed whether a peculiar wiggle in a graph represents an algorithm problem (such as neglecting to include a possibly negligible term) or a modeling one (such as ignoring a possibly important physical process). Usually it turns out to be nothing so esoteric: we come back the next week and learn that it was a bug.