One in a Krillion

Krillion Scoring: How Rarity and Answers Are Judged

How Krillion decides your answer is rare: familiarity data, manual passes, aliases, the depth zones, and why the too-clever pick sinks.

Two separate questions

Krillion is not scoring one thing, it is scoring two. First: is your answer inside the accepted answer set? Second: how rare is that answer inside the set? Almost every misunderstanding about the game comes from collapsing those two questions into one.

An answer can be true about the world and still be outside the set. An answer can be inside the set and still score almost nothing. And an answer can be rare without being the best pick in the round, which is the part that surprises most players.

How the answer set is built

Before a prompt can be scored, there has to be a list of acceptable answers. The developer builds that list by aggregating datasets — dictionaries, wiki pages, encyclopedias, whatever sources a given prompt requires — and then merging and cleaning the result into a single answer set.

This is why the accepted range is wider than you would guess. Some prompts accept thousands of answers, with aliases, alternate spellings and older names folded in. It is also why the occasional fair answer does not appear at all: the aggregation missed a source, and the developer says so plainly in the official FAQ.

The takeaway for a player is not to hunt for the set. It is to understand that the set is a body of data, not a single gold answer — and that a rare entry inside a large set is what a deep dive looks like.

How rarity actually gets decided

The official FAQ splits this into two cases, and the split matters for how you should play.

Some prompts can be scored mechanically, against a familiarity database of how often a word or a name actually turns up in what people read and write. That is a frequency measure, not an opinion. If a term appears constantly in ordinary text, a lot of players will reach for it, and the scoring treats it accordingly. There is no clever phrasing that rescues a common word.

The rest of the prompts cannot be reduced to a number that way. For those, the answer sets get a combination of language-model assistance and a manual pass. That is the honest answer to a question a lot of games dodge — yes, some rarity judgments involve machine assistance plus human review, not a pure formula.

What that means on the clock

Because most of the judgment is about how commonly a word appears, the useful mental model is a frequency ladder. At the top: the words a stranger would say in ten seconds. Below that: words a fan of the subject would say. Lower still: names a specialist uses, or older names for the same thing. Deep: things that exist in reference data but almost never come up in ordinary writing.

Twenty-five seconds is enough to walk down one or two rungs of that ladder. It is not enough to climb into territory you do not actually know. Which is why guessing a domain you have real knowledge of beats guessing a fancy-sounding word you just invented.

The depth zones, and the prize at the bottom

Your score is rendered as a dive. Shallow answers stay near the surface where light still reaches; rarer answers sink through the midnight zone, past the abyssal plain, down to the hadal zone at the trench floor. The best available answer in a round is called the Krillion.

The Krillion is not awarded on rarity alone, and the developer explains why. Past a certain depth, rarity gets hard to tell apart, and answers can get so obscure that virtually nobody would ever type them. So the gem is also picked for how satisfying the answer is to land on — it should be something at least a few players could realistically get.

The practical implication: the absolute most obscure thing you can think of is not automatically the best play. You want deep and reachable, not deep and unguessable.

The too-clever pick

The official FAQ gives the failure mode a name: too clever — the go-to obscure pick that is not actually that obscure.

Notice why it fails. It is not rare in the world; it is rare in conversation. It feels like a deep cut to the person saying it, which means it also feels like a deep cut to everyone else, and thousands of players type it at once. The game then prices it as what it is: a common answer wearing a disguise.

A workable filter: real deep cuts usually come from a narrow domain, not from an unusual word. A specific species, a minor character from a period, a local name, an older term — those are structurally rare. A fancier synonym for something everyone knows is not.

Aliases and autocorrect

Spelling is forgiving, and that is deliberate. With thousands of accepted answers and aliases, autocorrect checks alternate names too — so being a letter or two off can still resolve.

The official example shows the risk: the fish Threebanded Sheephead carries "sea donkey" as an alias, one letter from "sea monkey". A near miss can slot into an answer you never intended. You are scored, but on the wrong thing — a specific failure mode worth knowing when a round scores oddly.

The short version

Acceptance and rarity are separate judgments. Common words cannot be rescued by clever phrasing. Deep cuts come from narrow domains, not fancy synonyms. The Krillion rewards answers that are rare but still reachable. If you want to turn that into tactics, read rare answer strategy; if you have not started yet, begin with how to play.

Common questions

What does "the Krillion" mean?

The Krillion is the best available answer in a round. It is chosen for rarity, but not on rarity alone — past a certain depth the developer also weighs how satisfying an answer is to land on, so it stays gettable.

Why does an obvious answer score almost nothing?

Because it is measured against a familiarity database of how often words appear in what people read and write. A common word is common — the game expects most players to reach for it, and scores it accordingly.

Can a correct answer still be rejected?

Yes. A rejected answer usually means a source was missing from the aggregated answer set, which the developer acknowledges in the official FAQ. It is a gap in the data, not a ruling that you were wrong.

Does Krillion use AI to judge answers?

Partly. The developer's FAQ says prompts that cannot be scored mechanically against a familiarity database get a combination of language-model assistance and a manual pass.

What is a "too clever" answer?

The too-clever answer is the obscure pick that is not actually obscure — the first unusual thing most people would think of. It feels rare to the player and to everyone else at once, so it scores shallow.