I would agree: first attempt I scored 74 and analyzing their distance matrix it looks like their measure is broken. e.g. hammer and time are relatively close be

I would agree: first attempt I scored 74 and analyzing their distance matrix it looks like their measure is broken. e.g. hammer and time are relatively close because "hammer time" is a phrase so their proximity frequency is high, not because the words are semantically related. Second attempt was 89 after a trivial change to account for the broken measure. Nice idea for an experiment but the execution is lacking.