castlefall rules

How to play. Back to the game.

Castlefall (Castle of the Devil + Spyfall) is a party word game for usually six to ten people, preferably an even number, although it’s playable with a few more, somewhat playable with four or five, and playable in some theoretical sense with three. It was invented some time in 2017 by some folks at Epsilon Theta, an independent living group at MIT.

You can specify the name of the room you want to be in by appending a hash — /castlefall/#example joins the room called “example” — or change rooms with the “change room” button. There is no way to discover rooms with other players in them. Castlefall is meant to be played by people using the app but coordinating out of band, most often in the same room in real life. There is a basic chat feature, but the game is more fun in person.

Gameplay

Players load the page and join the same room. When somebody starts a round, all players receive a list of words randomly sampled from the larger word list chosen at the start. This list is the same for everyone in the round but shuffled differently on each player’s device. Each player also receives one of those words as “their word”. Players are randomly assigned to two equally-sized teams (or nearly equal, with an odd number of players), such that each player on a team has the same word and the two teams have different words — but nobody knows who else is on their team.

Players then communicate freely, trying to work out either (1) who else is on their team, or (2) what the other team’s word is, while not revealing their own word to the other team. You should not show other people your screen or try to see anyone else’s. If someone asks you to repeat what you’ve said, it is polite to do so as best you remember. It is also polite to avoid excessive cryptography.

This continues until somebody declares victory — usually announced by clapping loudly, since earlier declarations take precedence and there can be a race to be first. There are two ways to declare, which may or may not succeed:

1. Choose N players (including yourself) and claim they are all on your team. N is usually 3 for 6- to 8-player games and 4 for 9- and 10-player games, though players should adjust it to taste. For smaller games N can be “your entire team”, so you must name exactly the set of people on your team, including correctly guessing whether you are on the smaller or larger team in a 3- or 5-player game. Nobody else can declare victory with this method after you have done so. Start a one-minute timer and keep discussing; if nobody declares victory by method 2 before it elapses, the round ends, and you and your team win if and only if your declaration was correct.

   During that minute it is polite to repeat the clues that led to the declaration, so the victory comes from clever clues rather than from someone happening to overhear something others didn’t.

   Suggested N:

   • 6 or fewer players: N = the size of your team — you must claim exactly the correct team. With an odd number, you must also guess whether you are on the smaller or larger side.
   • 7 players: N = 3 for easier games; N = the size of your team for harder ones.
   • 8 players: N = 3.
   • 9 players: N = 4; or call 5 people including yourself, 4 of whom must be on your team.
   • 10 players: N = 4.

2. Guess the other team’s word. The round ends immediately, overriding any ongoing method-1 declaration; you and your team win if and only if the guess was correct.

   Even while a method-1 declaration is running, anybody may declare by method 2 — including people named in the declaration, and the original declarer themselves.

No matter who declared, which team they were on, which method they used, or whether it worked, you always win or lose together with your team — the set of people who had the same word as you.

The app does not judge victory declarations — it only records what the room agrees on. Once the teams are revealed (when the next round starts), anyone can click a team’s word on that round’s card to record it as the winner; while a round is still live you can instead type the winning word into the “declare winner” box, the same way you would declare victory by method 2. The app checks only that the word is one of that round’s two team words, then adds one win to everyone on that team and updates the leaderboard on the right of the page. Scores are per room, follow names, and are only as honest as the room. The timer, when started, appears as a bar across the top of the screen.

Strategy

The usual approach is to give clues about your word that are recognisable to somebody already holding that word, but not so obvious that opponents can pick it out of the seventeen-or-so other options. The whole game is that balance. You can react to clues you don’t actually recognise, to trick someone into thinking you’re on their team. You can also give clues about words you suspect belong to the other team and watch who reacts. That last one carries a risk: you may convince somebody that word is yours, and they may declare on it — and they may or may not be on your team.

Variants

Words can be replaced by any finite set of distinguishable objects, e.g. Dixit cards. That isn’t implemented, though you could use a word list of integers and map them to cards out of band.

Notes on this version

The app assumes people who trust each other to play honestly, so there are no access controls. Anybody in the room can kick anybody else, including themselves. Joining a room under a name already in use takes that name over — that is how you reconnect as yourself and keep playing after your device drops off.

Two things differ from the original. Autokick is off by default: locking a phone drops the connection, and with autokick on that would drop you from the room, so instead your name is held and marked disconnected until you come back. And the page reconnects on its own, with backoff, and again the moment you bring it back to the foreground — so a locked screen mid-round is not the end of your game.

← back to the game · home

Rules text adapted from the Castlefall README by betaveros (MIT).