I've written Clive/SimpleHanabi as an introduction to the kind of thing people might like to know during their first dozen games of Hanabi. And a guide to the kinds of things I expect to be telling new players. And to the kinds of things I feel it's prudent for experienced players to do around newcomers.
Then there is more advanced Hanabi. I could have written a document describing advanced Hanabi; I have not; this is not that document.
This is a document describing how to play Hanabi absolutely as well as I know how. As such, it is fairly uncompromising. The overarching principles are:
Nobody ever makes a mistake
There is no way to mitigate mistakes
There is no limit to how badly wrong things can go wrong if somebody makes a mistake
Whenever there are multiple well-defined options with clear meanings, the player will never choose a strictly worse one over a strictly better
Every negative inference that can be drawn, should be drawn
Playing less than optimally is a mistake (see above)
All players are perfect logicians at all times
All players can be relied upon to make all available inferences from every aspect of the game state at all times
All players should always act in the expectation all other players will do likewise
Not being a perfect logician, or expecting another player not to be a perfect logician, is a mistake (see above)
Any convention, no matter how rarely used or complicated, is justified provided
It gives a benefit
The risks are manageable
It doesn't actually result in cognitive overload for the players
Once agreed upon, no convention is optional
Negative inferences may be drawn when a convention isn't used
Failing to use a beneficial convention is a mistake (see above)
In the simple Hanabi document, I said Hanabi is an endlessly fascinating game. None of the above detracts from that: it just builds a new platform from which it can be even more fascinating. Nothing in this document detracts from the ability for players to make judgement calls, weigh alternatives, invent new techniques for dealing with new situations; think of it more as specialising and refining the language people speak by their play.
There is ample opportunity here for things to go pear-shaped. And for people to be forgiving when they do. (-8
Belief
There are seven levels of belief concerning some aspect of the game's state:
Fact: A thing definitely known
Nothing can ever be construed as contradicting a fact
Nobody can ever be construed as attempting to contradict a fact they know
Information: A thing the player should absolutely rely upon
Information is typically used at once, so there is no opportunity to contradict it
If it's wrong, a mis-play or unsafe discard can result
Inference: A thing definitively deducible from facts and information
Inferences will only exceptionally be false
Only a fact can contradict inferences
Inferences can and should be relied upon absolutely and indefinitely
Expectation: A thing a person should rely upon unless contradicted
Expectations can routinely be false
They are readily contradicted
But, like information, they can be relied upon absolutely unless contradicted
Suspicion: A thing that should sway a person between choice of otherwise similar merit
A suspicion will never cause a play or prohibit a discard, but might dictate which card to play or discard
Hunch: Things players believe or speculate
Players may act on hunches, but should be aware they're "off piste" when they do
Other players may be alert to people acting on hunches, but should assume they did not
Ignorance: Good luck. (-8
By way of illustration, suppose:
I know C is "yellow" and B,D,E are not.
b1,mc1,mc2 are in play.
I am told "A,B,C are blue"
My state of belief is therefore:
Fact: A is blue/mc, B is blue, C is mc, D/E are not blue/yellow/mc.
Information: A is b2 or mc3, and is playable.
Inference: That clue was more efficient than any of the alternatives
Expectation: If I play A, B will then be playable (b2 or b3, depending), and C after that (mc3 or mc4, depending)
I play A and it's the mc3. Before my next turn I'm told "B is a 4".
Now my state of belief is:
Fact: A is not a 4, B is blue 4, C is mc non-4, D/E are not blue/yellow/mc/4. Neither B nor C is playable.
Information: E is safely discardable.
Inference: One of C,D,E is a 3, or I'd have been told "A is a 3" instead of "A,B,C are blue"
Expectation: A is not playable, or I'd have been clued to that instead of "B is a 4".
Suspicion: B and C should not be discarded
Hunch: C is plausibly mc5. If it was the other mc3, maybe I'd have been told "A,C are 3" then "C is 3" instead to give me a discard. If it was mc2 or mc1, possibly "A,B,C are blue" then "C is 1" or "C is 2".
Hunch: Card D might be worthwhile in some way. I know E isn't, because I'm being allowed to discard it, but someone thought it worth getting me to play my mc3 now instead of waiting for me to discard E,D,C then cluing it more efficiently.
That's just on the basis of my own hand, of course. I may have more beliefs based on others' hands, the tableau and discards.