Answer lists
For many categories, candidate answers come from Wikidata, the open knowledge base maintained by the Wikimedia community and released under the CC0 public-domain dedication. For categories that need broad, everyday answers, such as things you would find in a classroom, we write the lists ourselves, with AI assistance. Every list, whatever its source, is reviewed with AI assistance, with errors removed and common alternative names and spellings added.
Word frequency
The word-rarity part of a score comes from wordfreq, an open-source Python library by Robyn Speer and contributors. Its data is released under CC BY-SA 4.0. We use it only to estimate how common a word is in everyday writing, and we switch it off for prompts about names or situations where it would not say anything useful.
Fonts, art and code
The text is set in Fraunces and DM Mono, both used under the SIL Open Font License 1.1 and hosted on our own server. The artwork (the plane, the clouds, the landscape) and the game code are original to this project.
Why open data
We use open data because a game that asks for the names of things should stand on shared, checkable knowledge. Wikidata is edited by volunteers across the world, and its records are linked to thousands of languages. That breadth helps us include answers that are not specific to one country. The trade-off is quality: volunteer-edited data is uneven, and a list pulled directly from it can contain oddities. This is why every list is reviewed and corrected.
What we change
- Remove things that are not what the prompt means: a festival in a list of rivers, hobby painters in a list of famous painters.
- Add alternative names, spellings and plurals so that reasonable answers are recognised.
- Rename awkward labels to the form most people would type.
- Block inappropriate words using a list shared by the build and the game.
- Rate how obvious each answer is, since open data has no notion of “first to mind”.
Corrections and licensing questions
If you maintain one of the sources we use and see something we have done wrong, or if you believe an item on our site infringes your rights, please contact us and we will look at it promptly.
How we check
- Automatic tests make sure every common answer is accepted, with plurals and typical typos.
- Every score band must have answers, and only the summit can score 100.
- A word list blocks inappropriate answers in both the build and the game.
- The top of every list, and unrated answers that look suspicious, are reviewed with AI assistance and corrected when players report problems.
Limits
Open data contains errors, and any list of answers is incomplete. Ratings of how obvious an answer is are our estimates and can be wrong. The names of people, places, brands and works appear only to identify them and belong to their owners. If you spot an error or an omission, please tell us. Read more about the method in rarity without a crowd and how a prompt gets made.