> On the downside they're trying anyway, confident that if their rules are
> just vague enough they're bound to help. I notice that they haven't tried
> the benchmark Professor Pullum implicitly offers, typing several pages of
> a major literary work (say, Moby Dick, or Pride and Prejudice) into this
> software with everything enabled and verifying that it flags none of the
> excellent prose as incorrect.
So lightproof is designed to be minimal and give ~zero false positives. But your idea is a good-one :-) What would be awesome, would be if you could get several of these smallish but representative classic texts, and create some unit tests in the lightproof module, such that we can ensure that not only are there no false positives now, but there will be none in future :-) patches most welcome (this is a volunteer project). The lightproof git repo is here: