How Clamber scores how rare an answer is

Every answer in Clamber lands in one of six tiers, from "everyone says it" to "almost nobody says it". The tier decides how many rungs you climb. This guide explains where the tiers come from, why they are built that way, and what happens when you type something we have not seen.

The problem with asking people

The obvious way to measure how rare an answer is would be to ask players. Count how many people typed "Nile" for a river in Africa, and how many typed something else. That works once a game has a very large audience. It fails for a small one, because for the first thousand players the numbers are noise, and it fails for a different reason later: a ranking based on player answers can be steered. A group of friends can agree to type the same odd answer and push it into the rare tiers.

So Clamber does not use its own players as the measuring stick. The rarity of every answer is fixed before the day starts, and it is the same for everyone. It is built from public data about how widely known each thing is.

Where the numbers come from

Each prompt, such as "Name a river in Africa", is turned into a precise list. The list of valid answers comes from Wikidata, the open database behind Wikipedia: every item that is a river and is located in Africa, for example. We drop items that are not real answers, such as disambiguation pages and lists, and we require that the item has an English Wikipedia article.

Then every answer on the list gets a popularity score, built from two signals about its English Wikipedia article:

• Pageviews. How many times people read the article over the last year. We use monthly counts and take a steady middle value instead of a total, so that a single burst of news does not turn an unknown into a household name.
• Incoming links. How many other Wikipedia articles point to it. This measures long-term prominence. A famous river is mentioned by hundreds of other pages, a minor one by a handful.

The two are combined so that neither can dominate. Pageviews alone would rank a country that happened to be in the news above a larger one, and links alone would favour old, well-edited articles. Together they track what people actually know pretty well.

From a score to a tier

Within one prompt, the answers are sorted from most popular to least popular, and the sorted list is cut into six tiers. The most popular few answers are the "Crowd" tier. The least popular ones are "Rarest". The cut points depend on how long the list is, so a short list of well-known answers still has all six tiers.

There are also safety rules on top. An answer whose article is read very often can never be placed in the two rarest tiers, however short the list. An answer whose article is hardly read can never be placed in the Crowd tier. These rules stop a short prompt from handing out a top score for something that is actually well known.

What counts as the same answer

People type the same thing in different ways: abbreviations, short names, leaving out a leading "The" or "Mount", the US or UK spelling. For each answer we build a list of accepted forms, using the names recorded in Wikidata plus common variants. The matching ignores small typos. Each variant is tested against the same matcher the game uses, and any variant that could be mistaken for a different answer is dropped, so one typed word never points to two different answers.

When you type something that is not on the list

No list is complete. If you type something the game does not recognise, it costs 2 seconds and you can try again. The game can also run an automatic check on the spot. If it finds a real answer for that prompt, you get credit, but we cap the tier at Rare. Only answers that were on the list before the day started can score at the very top. That keeps the top tier fair, and it stops a fresh answer from being worth more than a carefully ranked one.

Unrecognised answers are recorded along with the prompt they were typed for. Once a day, they are grouped and checked against Wikipedia and Wikidata. An answer that is another way of writing something already on the list becomes an alternative spelling. A real answer that belongs to the prompt's own category is added to the list, and its tier then comes from the usual process. Anything doubtful is set aside for a person to read. This is how the lists get better over time, and it is why we ask you to tell us when you think something is missing.

What this scoring cannot do

Being widely read on Wikipedia is not the same as being widely known by you. A local band might be unknown to the world and obvious to your town. The data is also based on the English Wikipedia, so it favours the English-speaking world. We think this is an honest trade for a score that is the same for everybody and cannot be talked up. If you spot something that looks wrong, tell us, and see about Clamber for how we handle reports.

Now you know what the tiers mean, read how to find the rare ones, or play today's climb.