If you want to add or edit any conventions, please make a proposal on Hanabi Central discord.
ReactorZero emphasizes simplicity and clarity.
We first cover stable and reactive clues in 3-player games then discuss general gameplay conventions then present conventions specific to 4/5/6-player games and variants.
ReactorZero has only two types of clues: stable and reactive.
Every clue given to the next player is stable.
We name the player whose turn it is Alice and the subsequent players Bob, Cathy, Donald, Emily.
ReactorZero uses the following terminology and principle to describe properties of cards.
Play reveal: Any stable clue that touches a card and makes it become playable-by-empathy tells Bob that he may play that card. No further meaning. If multiples cards are play-revealed and could be the same card, only the leftmost card is called to play.
Otherwise, the meaning of a stable clue depends on the clue type:
| Clue type | Meaning |
|---|---|
| Color (touching new cards) | Play leftmost newly-touched card |
| Rank (touching new cards) | In order of priority,
|
| Reclue with rank or color (touching no new cards) | Play leftmost touched card |
A locked player must not discard any non-trash card until they receive a safe action.
An unlocked player without a known or signaled action (play or discard) may discard their leftmost unclued card (their chop) instead of giving a clue.
If a card that was called to play is not immediately playable, the clue has no meaning beyond the information it provided to the hand (stall).
Every clue not given to the next player is reactive. (All of Alice's clues to anyone not Bob are reactive.)
For a reactive clue, the clue recipient is called the receiver; every player whose turn is after the clue-giver up to the recipient is called a (the) reactor.
The clue gets an action from the reactor(s) followed by an action from the receiver. When Bob takes that action, we say that Bob "reacts" to the reactive clue.
The type of action (discard or play) in both hands is determined by the type of clue:
| Clue type | Meaning |
|---|---|
| color | odd number of plays from the reactor and the receiver (1 discard + 1 play, or 1 play + 1 discard) |
| rank | even number of plays from the reactor and the receiver (2 plays, or rarely 2 discards) |
The value of a reactive rank clue is the numerical value of the rank clue. A rank 1 (respectively 2,3,4,5) clue has a value of 1 (respectively 2,3,4,5).
The value of a reactive color clue is red = 1, yellow = 2, green = 3, blue = 4, purple = 5. With additional colors, colors should be assigned missing values. If no colors are missing, start back from 1.
The actions (play or discard) will come from one of Bob's slots and one of Cathy's slots, respecting the equation
Clue value = Bob's slot + Cathy's slot (modulo 5)
Example:
The 5 in "modulo 5" corresponds to the number of cards in each hand.
In order for Bob to know which action he should take, Alice and Bob need to agree on which action Cathy should take. This is achieved by the following targeting priority.
Targeting priority in Cathy's hand:
For targeting priority purposes, the status of cards (playable, X-away, trash) is determined before the clue is given.
If the clue would make an unplayable card play or a critical card discard, the target is moved to the next possible target in the priority list.
This section applies at all player counts.
Play order: A player may play their playable-by-empathy and playable-by-signal cards in any order. It is recommended that they prioritize playing cards that could not have already been played.
Self-revoking: Play signals are not generally erased if the play-signaled card might have already been played. If it is deducible from the card's empathy that it must have been played, all players mark that card as trash.
Fixing play signals: Play-signaled cards, that will bomb, can be fixed by:
Discard order: A player with multiple trash or discard-signaled cards typically discards them from left to right. In the endgame, if a player does not discard from left to the right, the clue might be interpreted as a positional discard. In any other situation, a player may choose to discard another trash/discard-signaled card than the leftmost one, for example to keep an omni trash card.
Fixing discard signals: Discard notes can be revoked by a lock (stable or reactive). All discard signals in the player's hand are erased.
Free discard: Discard of non-trash cards without discard signals do not have any special meaning.
A card that is playable-by-good-touch is not expected to play on its own.
Hard information is any information that comes from the empathy of the cards.
Soft information also includes pink promise and the possible identities of a play-signaled card based on their empathy.
The meaning of a clue can only be overriden by hard information.
It is not expected that players deduce any information beyond soft information. These extra deductions form contextual information (targeting priority, elimination notes, not given clues, card saved, ...). Players should in general refrain from taking any action (play/discard) from contextual information.
It is Alice's job to save Bob's chop. Alice assumes that Bob might discard their chop (if Bob does not have a safe action) even if there is a critical card on Cathy's chop.
The endgame starts when the pace is strictly less than the number of players.
Positional discards are only possible in endgame and get the last possible player to play the matching slot. Even in endgame, assume first that Alice is discarding her leftmost locally known trash over a positional discard.
The players should try their best to avoid lines that do not get the maximum score. Reactive clues that lose on the spot should be reinterpreted to satisfy that condition. Playing a card from context might also be necessary.
Cards are targeted as if there would be more turns to play. If a hand has less than the maximal amount of cards, the slot numbers are counted from left to right starting at slot 1.
A reactor bombs (or discards a critical card): The next player must ignore the reaction and proceed as if they were locked for that turn. The entire reaction chain is off including the reactor.
The receiver bombs: The targeted card is not considered signaled.
A play-signaled player discards: Signal their card again if there is any doubt.
The receiver can delay their action by giving a clue or by making a different action. In this case, the targeted card in the receiver's hand is still considered signaled.
A reactor can only delay their action by giving a clue. If that happens,
When a player delays playing a card, all players must remember all playable possibilities for that card based on its empathy. This information is expected to be used later to determine if the card is still playable, became known trash or needs a fix.
4/5/6-player games are significantly harder than 3-player games. Players should familiarise themselves with the conventions in 3-player games before attempting 4/5/6-player games.
All the conventions for a 3-player game still apply for a 4/5/6-player game. Stable clues work exactly the same but some generalisations for reactive clues will need to be made. As before, any clue that is not given to Bob is a reactive clue, and all players from Bob up to and including the clue receiver are involved in the reaction. A reactive clue gives one action to each player in the reaction chain.
To determine whether to play or discard, players looks if a color or rank clue was given.
A color clue means that the reaction chain contains an odd number of plays.
A rank clue means that the reaction chain contains an even number of plays.
To determine their action slot, each player uses the following formula:
Sum of all action slots = Clue value (modulo the number of cards in each player's hand)
In 4/5-player games, there are 4 cards in each hand so the equation becomes:
Sum of all action slots = Clue value (modulo 4)
Each player involved in the reaction determines the target in the clue receiver's hand, then work backwards to determine the target in the hand of the player before that, and so on until they get to their own hand.
For targeting priority purposes, the status of cards (playable, X-away, trash) is determined before the clue is given.
Retargeting refers to changing the target to the next possible target according to the targeting priority. Retargeting is required if the current reaction chain would cause a player to bomb or discard a critical card. When this happens, the players must try to retarget on players earlier in the reaction chain first.
Forced targets: A forced target (connecting to an X-away target on a later player) has the highest priority (above playable cards to its left). Targeting it is thus not considered retargeting.
Blocked targets: A card play-signaled by the current clue prevents cards of the same suit from being play targets on earlier reactors. These blocked cards are treated as unplayable, and therefore automatically excluded from the targeting priority as play targets and thus are not preventing another card to their right from being the highest-priority play target.
Any retargeting on Cathy is allowed. On players beyond Cathy (Donald,Emily,...), it is only allowed to retarget from the highest-priority play target to the highest-priority discard target (can be a lock signal).
1-away: Nothing is played on the stacks. Donald holds r2 b2 x x. Cathy holds b1 x x x. Alice gives a green clue to Donald. The play target on Donald can only be r2. And Cathy does not hold r1 so the highest priority card on Cathy is b1. So Bob is playing r1 from slot 1. The is true even if Cathy knows that their slot 2 is not red.
2-away: Donald holds a 2-away blue card on the left and a 2-away green on the right (other cards are trash). Alice gives a color clue to Donald. Cathy holds the 1-away green card but not a 1-away blue card. The higher priority 2-away blue card prevents the 2-away green card from being targeted even if it is impossible for it to be the target from Bob's point of view.
2-away orange: Donald holds a 2-away orange card on the left and a 2-away green on the right (other cards are trash). Alice gives a color clue to Donald. The higher priority 2-away orange card prevents the 2-away green card from being targeted even if it impossible to get 2-away orange card to play with a color clue.
Mid-way retargeting: Occasionally, the target on a later player may change after an earlier player has played his cards (due to someone unexpectedly playing into a finesse). It is Alice's responsibility to make sure that the clue works. The intermediate players with a playable card must still target the x-away card even if it is known to them that a mid-way retargeting happend.
Play-discard-discard assumption: As an example, Donald has a one-away card, Cathy has no playable (and no empathy), and Alice gives a color clue to Donald. Whatever Bob plays, Cathy will likely play into Donald one-way card which would cause a bomb. In this case Bob has to trust that he holds the remaning copy(ies) of the connecting card, so that the clue works as play-discard-discard.
This section is only applicable when playing a game with special suits. For more on variants, see the variants page.
In variants including a pinkish variants (Omni, Pink, Light Pink, Chimneys, Funnels)
Pinkish special-fives includes Omni-Fives, Pink-Fives, Light Pink-Fives. All pink conventions applies but with a weaker pink promise. A clue focusing a 5 must be given by a rank 4 clue. Thus a pink promised 4 might be a 5. If all 2s are played, you can focus 5s using a rank 2 clue.
Pink conventions do not apply.
Stable clues function the same as in normal reactor except that:
The meaning of reactive clue changes:
To be able to get more reactive clues (at the cost of a weaker color promise), we regroup colors in buckets: (red-yellow), (green-blue), (purple-teal). In special variants, teal is replaced by a special suit, so the last bucket is (purple-special suit).
Bucket order: (red-yellow) -> (green-blue) -> (purple-teal) -> (red-yellow).
Bucket promise: The buckets of play-signaled cards from a rank clues (resp. color clues) must be increasing by one (resp. decreasing by one) with respect to the bucket order.
Bucket lie: Reactive clues can lie on the promised bucket, this is not a reason for retargeting. In case of a known lie, we use the bucket promise priority: finesse > same bucket as the previous player > remaining bucket. Promising a bucket that is forbidden or completed is considered a lie.
Superposition: A player plays a playable card with a bucket a promise and without enough empathy for the color of the card to be known. After playing, the player is said to be superposed for that bucket.
Bucket restrictions: some superposed buckets cannot be targeted and promised.
Targeting Cathy: Cards from superposed buckets from Alice or Bob cannot be targeted on Cathy. The forbidden buckets also are skipped in the bucket promise.
Play-signalling Bob : Cards from superposed buckets from Alice cannot be reactively play-signaled by Alice on Bob. The forbidden buckets also are skipped in the bucket promise.
Collapse: It is important to agree on which buckets are superposed for which player to be able to correctly target Cathy and play-signal Bob. We conventionally when a superposition collapses, which might happen much later than a contextual collapse.
Immediate collapse: The global empathy of the play-signaled card is enough to determine if their card is of the first or second color. The player does not enter superposition is they were not already in superposition.
Visible collapse: The superposition for a player collapses when they see another player play a card from their superposed bucket.
Trust collapse: Bob is asked to play (in a stable or reactive fahsion) a possibly playable card from a superposed bucket for Bob. This is a called to play on that card. If Bob can infer its identity from its empathy, superposition collapses as soon as Bob plays it.
Lie fix: Bob is asked to play (in a stable or reactive fashion) an unplayable one-away card from an unsuperposed bucket for Bob. This is called to play on that card. Bob should think they played the connecting card earlier instead of another card and erases the lie from their stacks if they can deduce it.
Asking collapse: Alice is superposed for a bucket. Any stable colour clue to Bob within this bucket promises the lowest possible playable rank of that suit from Alice's point of view. The superposition globally collapses as soon as Bob plays/discards the asked card. Bob discarding may not collapse the superposition fully.
The sum of the color values (or bucket numbers) of the played cards must equal 1 for rank and 2 for colors (modulo the number of colors/buckets).
Knowledge of these conventions is not expected.
These moves can usually performed even if one of the player is not aware of them.
These moves should only perform in desperate situations.
These conventions must be explicitly turned on before the game starts.
2026/07/16: Rework explanations for games with 4 players or more.
2026/07/14: Remove negative trash reveals. Trash reveals must now touch the card they are revealing.
2026/07/13: Remove fix clues. In other words, fix clues keep their stable meaning.
2026/05/20: Move max play re-targeting to common unconventional conventions and restrict it to the end game.
2026/05/20: Update discard re-targeting convention to prioritize Cathy's playable over Donald's leftmost one-away.
2026/05/16: Added the optional convention of Orange Trash Finesse Position Exception for orange variant.
2026/03/21: After an "unconventional" bluff, the reciever with an unplayable play-signaled should proceed as if they were locked for that turn.
2026/03/21: Define cases where midway re-targeting happens and what the middle player(s) should do.
2026/03/21: The meaning of all stable clues are now to be interpreted from the shared point of view of Alice and Bob.
2026/03/21: Allow again negative trash reveal with rank touching new cards.
2026/03/15: Swap priority order of locks and trash reveals to make possible to revoke discards by locking even if the clue reveals trash.
2026/03/15: All cards signaled in a later player's hand by the same clue including away cards, are now considered trash for the purpose of the clue.
2026/03/09: Trash reveals now only applies on trash cards that are newly revealed. Trash reveals must now touch the card they are revealing.
2026/03/08: Reactive locks, non-revokable discards except by a lock and no skipping of obviously playable.
2026/03/07: Prioritizing playing cards that could not have been duped.
2026/03/02: Reciever bombs revokes the signal given by the latest clue.
2026/03/02: Definition of the focus of a clue for pinkish variants. Simplified/stricter pink promise for pinkish fives. A clue focusing a 5 must be given by a rank 4 clue.
2026/02/18: Reciever bombing or any player discarding does not revoke play-signals in their hand anymore.
2026/02/07: Changes to the orange conventions. Discard-signaled or playable-by-empathy orange cards (known to be onlsignalsy orange by empathy) are obviously playable and cannot be revoked. Duplicated play-signaled cards are assumed to be not orange. Play-signaled orange cards (known to be only orange by empathy) get revoked.
2026/02/05: Pink promise reclue known lies are now play clues.
2026/01/25: Pink promise now also applies to playable rank clues when the focused card is a trash rank (called to play for play clues or locked card). Discard clues may not be given with a trash rank anymore.
2026/01/25: For simplicity, Orange cards are never considered obviously playable, including the ones that are playable-by-empathy.
2026/01/25: Drawing a new card revokes discard signals in that hand.
2026/01/11: An action by the reciever explainable only by a wrong interpretation of the reactive clue erases the reactive signal from that clue.
2026/01/11: Discarding with a play-signaled card remove play-sletignals on that player for targeting purposes. They are still expected to play on their own if the discard was not an obvious mistake.
2025/12/08: Unconventional moves disclaimer: bluffs, shouts, endgame rank positional.
2025/11/29: Allow re-targeting over X-away targets blocked by a targeted card of the same suit on a later player within the same clue.
2025/11/26: New stable clue flowchart that is native and does not require login
2025/11/15: Change to Synesthesia convention allowing stable color clues.
2025/11/13: Discarding a critical card has now the same effect as a bomb for recovery purposes.
2025/11/09: Pink play lies are stalls (if deducible from "global" empathy).
2025/10/31: Pink promise conventions do not apply anymore to pinkish special-ones and special-fives. Playable rank for pinkish special-fives.
2025/10/29: Responsibility for remembering possibilities for a play-signaled card that was delayed.
2025/10/29: Add convention for bombing receiver.
2025/10/22: Renamed playability categories
2025/10/21: Change how playable rank clues work with an inverted suit.
2025/10/21: Change how playable rank clues work with special pinkish suits.
2025/10/20: Add special conventions for Inverted and Synesthesia.
2025/10/07: When fixing a play-signaled card with a stable clue, the meaning of the stable clue is now ignored.
2025/10/01: Added suggested learning path.
2025/09/26: Remove Maximum Play exception to Re-targeting.
2025/09/26: Revert to simple targeting priority in the receiver's hand. Cathy's play does not have priority over B-D finesse anymore.
2025/09/25: Simplify playable dupe exception: playable dupe considered playable unless it has just been played signaled in a later player's hand as part of the reactive clue.
2025/09/21: Change to targeting priority: playable-by-signal cards are now prioritized over X-away cards in all hands. Remove Inbetween priority.
2025/09/17: Add Maximum Play exception to Re-targeting.
2025/09/12: Define dupe and introduce different-hand dupe priority for 3-player games, to be consistent with games with more players.
2025/09/10: Add No Context and Best Effort sections for clarification, as well as more optional conventionsChat
2025/09/07: Remove Connecting Principle and add No Finesse Re-targeting
2025/08/30: Update Connecting Principle
2025/08/25: Update Delay Convention, Inbetween Priority and Connecting Principle
2025/08/24: Define which cards are play-signaled.
2025/08/24: Play reveals that are not global do not override the meaning of a stable clue anymore. For example, a playable rank clue locally revealing a card still gets the leftmost card.
2025/08/23: Simplfied definition of global knownledge.
2025/08/22: Conventions for 5-player
2025/08/22: Add Cathy's playable priority levels in 4-player priority table.
2025/08/21: Make "play reveal" have top priority.
2025/08/20: Locks now have priority over playable rank (in pinkish variants)
2025/08/17: Use highest before leftmost as a tie-breaker for sacrifice.
2025/08/17: Conventions for 4-player (target playable-by-signal before finesses in Cathy's hand, different hand dupe, cathy connecting principle)
2025/08/14: Playable rank reformulated as good-touched rank.
2025/07/19: Remove different hand dupe for 4-player or more.
2025/07/17: Added playable rank clues.
2025/07/05: Different hand dupe priority for 4-player or more.
2025/07/05: Emily’s priority for 5-player games
2025/07/05: Simplified finesses in 4-player games
2025/07/04: Add pink promise for play clues
2025/07/04: Change back order of discard from left to right
2025/07/04: Remove queued plays (playable card must be immediately playable and can be played in any order)
2025/07/04: Use playable reveals as stable clues.
2025/07/04: Change stable discard to ref sieve style.
2025/07/04: Move leftmost playable-by-signal above leftmost trash in the priority list.
2025/07/04: Change order of discards to right to left.
2025/07/04: Add pink promise on locks.
2025/07/04: Add endgame positional discards.
2025/07/04: Change stable clues to direct stable clues.
2025/07/04: Add stable clues from referential sieve.
2025/07/03: Bob's delaying revokes signal to later players in the reaction.
2025/07/02: Document cleanup
2025/07/02: Remove finesses from trash-reactive targeting priority.
2025/07/01: Added the no good touch principle.
2025/07/01: Remove slot 1 exception for finesses and rules for finessable dupes.
2025/07/01: Initial document.