This code is only fast in the 'happy' case (ie. ~400 unique places, with no odd distributions, temperatures with a std dev of 10 from a mean between 0 and 40).

This code is only fast in the 'happy' case (ie. ~400 unique places, with no odd distributions, temperatures with a std dev of 10 from a mean between 0 and 40). It is still correct in the 'unusual' cases, but would be slow because it would revert to fallback code.