This is pretty neat. I read “No 3 slides.” They noted that, while testing has combinatorial explosion, most software failures in the field just take 1-2 interactions with it dropping off as interactions go up. Six interactions covers all but rarest. So, you can theoretically knock out most errors just testing for all n-way interactions where n is 1-6. They also have an efficient method and give away the software for free. Worth trying on test cases or various known-bad source (eg old versions).