[mbreese]

Given enough of a buffer and a historical corpus, it should be possible to do this in near-realtime, but likely with a few minutes of buffer. Especially with some kind of historical trained classifier, it should be possible to do this in close-enough-to-realtime-to-be-useful scale.