How Krillion Scores Answers | Rarity Explained

Krillion's whole pitch is that rare answers sink you deeper, but how does the game actually decide what counts as rare? It's the thing I kept wondering after my first few dives, and the developer's FAQ actually answers it pretty directly. Here's the scoring system with the guesswork taken out.

Rarity is the whole game

Unlike a normal trivia game where you're right or wrong, Krillion scores on how rare your correct answer is. Seven prompts, twenty five seconds each, same set for everyone, and the player who lands the rarest correct answer sinks deepest. An obvious answer is technically correct but nearly worthless, and the entire skill of the game is reaching past the obvious thing into an answer few people will type. If you're new, the rules page covers the basic loop; this page is about how the rarity score behind it actually gets calculated.

How the game decides what's rare

The developer's FAQ splits this into two cases, and it's worth knowing which one a prompt falls into because they reward different instincts.

Mechanical scoring: the familiarity database

Some prompts can be scored entirely by machine against a familiarity database, a measure of how often a word or name actually turns up in what people read and write. A common word like "dog" scores low no matter how right it is, while a specific species or a niche proper noun scores high because it barely appears. These are the prompts where specific, real, slightly obscure answers win, because the database doesn't care about your reasoning, only about frequency.

LLM-assisted scoring: the human pass

Other prompts can't be reduced to a number like that. The developer says those get a combination of LLM assistance and a manual pass. So yes, some of the rarity judgment is done with an AI model, but a human checks it, and this is where the "is Krillion AI" question usually comes from. The game itself isn't generated by AI, it's a solo developer's project, but part of the rarity scoring leans on an LLM for prompts that don't map cleanly onto word frequency.

Aliases and generous autocorrect

Krillion accepts a lot of aliases because people call the same thing different names, and the autocorrect checks those too. The developer's example is Threebanded Sheephead, which has "sea donkey" as an alias, one letter off "sea monkey" even though the two fish look nothing alike. This matters for scoring because a lesser known alias can be far rarer than the common name, so knowing a regional or niche name for something is a real way to sink deeper. The answers guide goes into how to use this on purpose.

Why the daily gem isn't always the rarest answer

This one surprised me. The daily Krillion, the gem answer, isn't chosen on rarity alone. The developer says that past a certain depth rarity gets hard to disambiguate, so the gem is also picked for fun factor, the answer that's most satisfying to land on. Some of the collected answers are so obscure that maybe one player would ever type them, and a gem that only one person can reach defeats the point, so it should be something at least a few of you actually get. That means the absolute rarest correct answer on a given day might not be the gem, which is a subtle but important thing to know if you're chasing the top spot.

Is there an Elo rating?

No. Krillion doesn't use an Elo system the way chess or some Wordle spinoffs do, where your rating moves based on your past performance against other players. The scoring here is per answer, not per player: each prompt's rarity is judged against the familiarity database or an LLM pass, independent of your history. You don't have a persistent skill number, you just have how deep each of your answers lands today.

The "too clever" trap

The FAQ names the biggest scoring mistake directly: "too clever" is the go to obscure pick that isn't actually that obscure. You drop the obvious answer, reach for what feels like a deep cut, but so does everyone else, which makes it common. A rare answer has to be rare for everyone, not just rare relative to your first instinct. That's the whole reason the answers guide tells you to move past your first and second responses entirely.

Bottom line: rarity is scored against how often an answer actually appears in the wild, partly by machine and partly with an LLM and a human check. So the winning move is always the same: a correct, specific, genuinely uncommon answer. For the mental habit that gets you there, see the answers and rare strategy guide.

← Krillion rules · Rare answer strategy →