I think its matter of "least common denominator" as in the sum of all fields will surely be unique, but what's the _minimum_ number of fields needed to isolate one user? You can download the JSON from each test and compare the diffs yourself - there's a lot of noise from "cpt" and "ratio" fields, but some that stand out are "referer" and "cookie" fields as well as a few SSL attributes. Not sure if controlling for those is all it takes to de-anonymize, but either way it's not great.