How the rarity score works
Short answer: each answer carries a score from 10 to 100, derived from how widely that subject is documented in public reference sources. Widely documented subjects are the ones most players think of first, so they score low. Thinly documented subjects score high.
Why not use real player data
The honest answer is that we do not have it yet, and we are not going to fake it. Games in this genre usually claim their rarity numbers come from millions of player answers. Some do; most started by estimating and never updated the label. We start with an estimate, we say it is an estimate, and we replace it with real data as rounds are played.
That distinction matters more than it sounds. If a game shows you a number and calls it “how rare your answer was”, you have no way to know whether it means anything. Every score on this site is tagged with its source, and the tag is visible on the category pages.
The method
- Build the answer list. Categories are assembled from structured reference data, so every word in the list has a documented subject behind it rather than being invented by a language model.
- Measure how widely each subject is looked up. We take public page-view counts for each subject over a rolling window, which is a free, stable proxy for “how many people already have this word on the tip of their tongue”.
- Convert to a log scale. Popularity is not linear — the gap between the most-viewed subject and the median one is several orders of magnitude. We take the logarithm so the top of the list does not flatten everything beneath it.
- Bucket into five tiers by within-category percentile. Scores are always relative to the category. “Rare” in a category of 200 household objects means something different from “rare” in a category of obscure species, and the percentiles keep that honest.
- Human review on the edges. Automated scores get a specific class of thing wrong, discussed below. Flagged entries are corrected by hand and marked as curated.
What this method gets wrong
One failure mode is worth stating plainly, because it is the most common criticism of every game in this genre and we would rather publish it than be caught by it.
Documentation volume measures fame, not availability in memory. A celebrity animal with a viral reputation can have enormous page views and still be one of the first rare answers a player thinks of. The score will call it common when, in game terms, it plays as rare. There is no way to detect this from reference data alone — it needs actual player answers, which is precisely why we keep a manual override field and why the override is the thing we most want to replace with real telemetry.
A second, smaller failure mode: regional vocabulary. A word that is obvious in Lagos and unknown in Leeds gets scored by its global documentation volume, which may sit in the middle when it is actually impossible for most players on either side. The list only accepts English-language names, which makes this worse rather than better: if your first instinct is a local name, the game will not recognise it at all. That is a real limitation of this version, not a scoring quirk.
Spelling and plurals
The game normalises case, spacing, hyphens and apostrophes, and it accepts singular or plural for most entries. It does not accept misspellings. That is a deliberate trade: forgiving spelling would require judging “close enough” on every answer, and a wrong judgement in either direction is worse than a strict rule the player can learn once.
What we do with answers we do not recognise
If you type something valid that the game does not have, it is logged rather than discarded. Those logs are the primary input for expanding categories, and they surface regional vocabulary that reference data misses.