Version 0.6
If you’ve found this document, you already know Hanabi and are familiar with some conventions. Most likely, you were introduced to a system in which “a clue means play this card”, and with plenty of other established strategies and contextual protocols. Follow the Leader is a new set of conventions for playing Hanabi based on the idea that “a clue establishes a relationship”. What the heck does that even mean? Read on to find out.
If you need a reason to try FtL, here are a few:
The first two protocols provide the foundation for everything else, and are sufficient to convey the gist of the system. So, here we’ll introduce two ideas: The Relational Clue Protocol and The Copycat Protocol. Later, we’ll go into greater detail and introduce more protocols.
| ★ Relational Clue Protocol ★ A clue establishes a positional relationship for the players NOT receiving the clue to use -- the leader and the follower. ★ Copycat Protocol ★ The follower copies the action of the leader -- play or discard -- shifted by the number of slots indicated by the relational clue. |
|---|
The key to the game is figuring out what the clue-giver intended for the team to do with the indicated relationship.
Example (Turn 1):
For ease of reference, the cards in hand are numbered from left to right: Slot 1 - Slot 2 - Slot 3 - Slot 4
In examples these refer to players A,B,C and D respectively, seated in that order, where player A (Alice) is usually the one giving a clue.
Any card in hand that would be a legal play if placed on the stacks when the player next acts.
Any card in hand that can never be playable (due to another copy already being played, or a lost suit).
A card in hand whose identity is globally known, i.e. everyone knows that everyone knows exactly which card it is.
A card can sometimes be Known Discardable or Known Playable, by deductive reasoning with only partial information. In this document, these are treated the same as Known cards, but they are not Known cards.
A card in hand with globally known instructions to play or discard.
A card in the Follower’s hand that allows everyone else to be on the same page. Usually it’s the only Playable card or only Trash card in a player’s hand, although prior information and context can enable other types of legal target. (See: Target Priority)
A Relational Clue is a clue that establishes a positional relationship for the leader and the follower.
A clue of GREEN or THREE(3) indicates the Clue-giver wants the Leader and Follower to perform the same action in the same slot in their respective hands.
A clue of YELLOW or TWO(2) indicates the Follower should copy whatever the Leader does, but one slot to the left. (BLUE or FOUR(4) indicates to follow one slot to the right.)
A clue of RED or ONE(1) indicates the Follower should copy whatever the Leader does, but two slots to the left. (PURPLE or FIVE(5) indicates to follow two slots to the right, which in four card hands refers to the same card.)
Relationships wrap around, so Slot 4 is left of Slot 1, and Slot 1 is right of Slot 4.
![][image1]
| CLUE | RED | YELLOW | GREEN | BLUE | PURPLE |
|---|---|---|---|---|---|
| CLUE | ONE | TWO | THREE | FOUR | FIVE |
| MEANING | 2 Slots Left | 1 Slot Left | Same Slot | 1 Slot Right | 2 Slots Right |
Players must determine, from context, whether a relational clue is meant as a play clue or a discard clue. Usually, this is determined by the Leader who assesses viable targets in the Follower’s hand. The Leader’s action, in turn, relays a “play” or “discard” instruction to the Follower.
The Target of a clue is the card in the Follower’s hand that allows everyone else to be on the same page. Agreement about what makes a viable target in someone's hand is an essential part of making FtL work.
To determine the target card in the follower’s hand, use the following criteria. If any cards qualify, the leftmost is the target. Otherwise, continue down the list until some cards qualify.
More advanced play involves incorporating more contextual information. The targeting protocol for advanced players is an alternative set of priorities that improve the chance for a team win.
Unknown playable
Unknown Trash
Known playable
Known trash
Non-critical
Critical
At certain points clues need to be given to cards to either discard them or play them directly. This mostly occurs in the late-game or when the board is configured in an unpleasant way. There are at least two situations where this will occur:
When multiple cards are touched by a direct clue, the focus of the clue is the card that makes most sense, and in case of a tie the leftmost of them.
It is possible that with enough contextual information a clue can be both direct and relational, but that is left as an exercise to the reader.
Never leave a player with 0-clues and no known play or discard.
In other words, any clue that takes the team to 0 clues MUST be a discard clue that regains clues immediately, unless there is a way to recover clues without discarding cards at random.
For example, if Donald has a card that is known to be trash, Alice can give the last clue as a play clue to Bob and Cathy.
A group may choose to toggle these special conventions “on” or “off”.
(Normally, one-away from playable cards are not viable targets.)
A card that is one away from playable may be a viable play target in the following two cases:
Conversely, one-away targets are not viable if asked to lead with a completely blind card.
(Normally, a leader can be double assigned such that a single act instructs two different followers.)
With role reversal, if a person is asked to lead twice, they are instead a follower for the second clue.
Example: Alice clues Bob, establishing a relational clue between Cathy and Donald, where Cathy leads. On Bob’s turn Bob clues Alice, asking Cathy to lead into Donald a second time. In this case Donald is now the leader for a clue on Cathy instead.
Take note that this relationship is established before Cathy plays a card, so everyone should take care to consider the relationship at the time of cluing.
(Normally, there is no preference for which clue to give if it conveys the same relationship.)
With fill-in priority, full information on a card must be filled in before issuing a same-relation clue that does not fill-in.
This yields significant deductive power, not limited to the following examples:
If a clue does not fill in the identity of a partially clued card, negative information is conveyed about that card. e.g. A 4 is touched in a player’s hand. If a red clue is given to that player, the 4 is known not to be purple as well, because giving purple would convey the same relationship and also fill-in the identity of the 4 card.
Also, if cards are touched by the same clue twice, they can be deduced not to have the corresponding identity. e.g. A clue touches some green cards in hand. Later, when the cards are clued as green again they are certainly not 3.
Leader(Bob) and follower(Donald) play same card, Observer(Cathy) fixes
A clue that took the team 0 clues should be treated as normal by the leader if the leftmost card with the least amount of positive information touched by it is trash. That card is promised trash.
A rank clue means that the leader should do the opposite thing of the action they want to assign to the follower
The assigning of the leader skips over players with known playables. If because of this rule no-one would be assigned a leader, ignore it for that clue.
If Cathy is the observer Bob can tell Donald to play an 1-away card. Cathy should play their left-most card that can make the followers card playable.
If Bob receives a clue that cannot be a direct play clue
These are conventions for play that follow from the foundational Protocols and the rules of Hanabi. It may be possible to infer them from what’s been previously established, but I state them explicitly here because I think they are mostly necessary to make FtL playable.
For the most part, ignore partial information for the purposes of determining a target. Instead consider only known and committed information.
Whenever a Leader plays or discards, the correspondingly positioned card in the Follower’s hand is considered “committed.” The Follower has the option to do something else instead of copying the Leader, but the committed card should be assumed by all to be played or discarded at the next opportunity.
A player currently expected to fulfill the role of Follower may simultaneously be called on to be a new Leader. Said player must perform the action demanded of the new Leader instead of playing or discarding the previously committed card.
The Leader should interpret a clue with as few assumptions as possible about unknown cards in their own hand. Known and committed cards in the Leader’s hand always get priority over unknown cards.
Importantly, cards with partial information do not get prioritized -- except when contextual information is sufficient to draw that conclusion.
Sometimes you just have to be resigned to give a useless clue because any other action would conflict with players' existing obligations.
Sometimes, a clue is confirmed to be a discard clue from context. (e.g. The 0-Clue Mandate).
In this case, a known playable card can still be played by the Leader, if it leads to the correct Slot to induce a good discard in the Follower’s hand.
On the other hand, if there are no targetable discards in the Follower’s hand, but there is a targetable known play, the Leader is expected to discard the corresponding Slot. The Leader’s discard does not override the established expectation for the Follower to play that card.
Strikes are a resource. Sometimes it can make sense to intentionally incur a strike to get a particularly pesky card to play. When the Leader misplays a card, always assume the clue-giver intended the misplay in order to induce a correct play from the Follower. Don’t forego The Copycat Protocol just because of a misplay.
Everyone has different tendencies. Does Alice tend to choose more straightforward clues? Does Bob often shirk his commitments to do other things? Does Cathy always discard her oldest card? Does Donald always try to manipulate who will be the next clue-giver?
As you play more and more, you’ll learn the tendencies of your teammates. These tendencies, if reliable, can be used to make more complex predictions and issue more elaborate commands.
This document does not demand optimal play, by design. I want everybody to play on their own terms. As long as you act within the Protocols and Principles of the system, you have freedom and options, and it’s not wrong to choose a suboptimal action.
FtL is taxing on memory and attention. Taking notes on committed cards helps to not lose track of them.
On hanab.live with H-group highlighting:
I use ‘cm’ for committed plays.
I use ‘kt’ for committed discards.
You’re right. They are confusing -- especially because when the Follower is expected to play the right of the Leader, the Leader is expected to play to the left of the Follower.
It may help to think of relationships moving forward in time:
It takes some getting used to, but after you’ve done it a few times, you’ll get the hang of it.
Redundant clues are fine -- often necessary, even. Efficiency lost to redundant clues should be easily recovered by relational clues throughout the game.
You can almost always issue a direct clue to the next player to act (Bob). When the observer sees that the clue would fail as a relational clue, they are obliged to treat it as a direct clue.
The optional tool Role Reversal further improves possible clues to give.
As always with this system, figure out what’s best from context. If it’s not possible to figure it out from context, your teammates should be promising whatever interpretation you go with is fine.
A reasonable use of this ambiguity would be when both clue-touched cards are duplicates of each other. Either one could be played or discarded and the other deduced by elimination note.
The 0-clue Mandate is supposed to be used to avoid this situation, but mistakes happen. The situation is undefined, so there is no conventional expectation.
Faced with such a scenario, I would consider if partial information or historical context about any of my cards was enough to warrant a play or discard.
If the team agrees to allow them beforehand, or if you are very careful about context, finesses and prompts should work, inducing an unexpected play from the observer . I do not usually consider one-away from playable cards to be clear targets, so finesses demand the utmost contextual care.
I've got nothing for ya'.
For 3 players, ignore the observer role. Any clue talks to the other two players as leader and follower, regardless of whose hand the clue touches. Here’s an example. Here’s another. And another!
One way to play with 5 players is to treat the player preceding the observer as the follower and the other two unclued players as potential leaders. Here’s an example game.
I have yet to imagine how a 2-player game might be possible. As for 6 or 7-player games, wow! You have way more friends than I do. 6p
I have yet to think about them at all, but I imagine some variants should be impossible. I also suspect some variants could be way easier with this system.
Omni, rainbow, and pink suits come to mind as easier. Null, white, and brown suits should be harder.
I wrote about it in the next section: Inspiration and Goals.
Hey, isn’t this like Hat-Guessing?
There are some similarities, however this convention system is a lot easier to use and allows for more freedom in the clues you can give.
In my return to playing Hanabi after several years’ hiatus, I was thrilled to find a flourishing diversity of convention sets that really smart creative people have devised. Follow the Leader is a way to play that suits my own sensibility, inspired by other people’s developments.
It’s in my nature to break things down to elementary parts. So when we all learned some of the conventions of Hanabi, my instinct was to break it down to the fundamental assumptions and agreements of play. Even with very few assumptions, quite sophisticated maneuvering is possible.
The essential foundation for most of the traditional conventions, in my assessment, is: A clue means “play this card.” Assuming we’re trying to win within the rules of the game, even just the one simple assertion implies a ton, yielding strategies like Finesses and Good Touch Principle.
I got to pondering what else could a clue mean? I kicked around some silly ideas like: what if a clue means sort your hand. It was cool to learn what others have come up with in this regard. Referential Clues mean “play the card next to the clue” and were a revelation. Hat-Guessing is a different monster whereby a clue means whatever the lookup table says it means.
I became enamoured with what I call Q&A play in our conventions, where one player poses a question with a game action (clue/play/discard) and other players answer their question with a game action of their own. This happens most commonly with finesses and bluffs. You can know the true identity of your clued card if another player plays blindly. It’s a fascinating means of communication.
Follow the Leader is a set of protocols that relies heavily on Q&A play and builds off of a simple idea: What if a clue means “this is an interesting positional relationship that should be used to convey further information.” It’s a leap from its predecessors, but it works surprisingly well.
There are innumerable ways to “win” Hanabi, but I have my own priorities for play that make some convention sets more enjoyable than others FOR ME. These are my priorities -- the things I love about Hanabi -- and how I try to reinforce my priorities with FtL.
Granted, these are all things the base game of Hanabi achieves with flying colours. What can I say? Bauza’s a brilliant designer.
The following are links to hanab.live replays. They were created during development and do not accurately represent the current version of this convention document. Current, correct replays will be added later.
Example Example Example Example Example (this one has a tricky start)
Example Example Good Example, but a loss Example
Example Example Example Example Example
Example Example Example Example Example
Example (some tricky stuff with dealing with one player’s non-trash hand, but being forced to give a discard clue -- context maybe strong enough to avoid misplays… maybe not)
Much respect and gratitude for how these people have advanced the art of Hanabi play and helped develop Follow the Leader.
H-Group
Referential Sieve
Hat Guessing
RavenGr
KS, NJ, HM, YJ, MS
tODDlife, Planky, Broken, Dfogo
Doodles, Timotree
ravengr, antitelharsic, paula, & arv
[This page is intentionally left blank.]