As an example - when I have to retest a wafer, I sometimes find that a few devices which had passed the first time, fail at retest, or vice-versa.

What I needed was a tool which 'merged' two input files, and produced a single output file which, for each serial number, contained the 'best' result from each of the input files.

... I can picture the scenario now: 'So, if the yield was 90% both times you tested the wafer - how come you now claim it's 95% ?'

Hmm. I suppose I'd be reluctant to use this approach to summarizing failure rates as well. Testing circuits on wafers is way outside my field, but I would have expected that if there is a subset "A" that fails on one pass, and a subset "B" that fails on another pass, then the set of troublesome serial numbers to report as unreliable should be the union of sets A and B, rather than their intersection.

I can understand the perspective that the only "real" failures are the ones that consistently failed on every pass. But there is the other perspective: that the only "real" successes are the ones that never failed on any pass. Fortunately, perl makes it easy to report the results, no matter which perspective you choose.