Discover, 10 minutes before you were planning to go home, that at some point while you were getting your data files ready, you typed "x=60" instead of "x=70", and that for all of the tests of the analysis you've been working on, you've been comparing model data from some random cornfield in Iowa with observational data from a different random cornfield in Nebraska, 350 miles to the west.
Happily, in pretty short order I was able to regenerate the datafiles and run everything through the analysis again, and it all still works. In fact, it works even better when you use the right data, which is not really surprising.
And it's not like I had actually published anything on this yet, either, so no harm done.
But I was sweatin' bullets for a little while, there.