It lives in one trait, and only one. The phrase on every token is set in one of 16 typefaces, and those are not handed out evenly. Everything else is.

50 symbols multiplied by 40 phrases is exactly 2,000, and every pair exists once, so every symbol appears exactly 40 times and every phrase exactly 50 times. No symbol is rarer than another and no phrase is either. You can count it in the metadata as published.

So a token's rarity is its font's share of the supply, and nothing else. There is no score, no composite rank and no weighting we invented. The counts are below.

sort --by font | uniq -c