[observation]

I realize this contradicts a core CompSci idea of bad data in, bad data out, but if I recall right then at the higher levels of rationality like neural networks we have similar counterintuitive results.