Empty clues have no logic

[Silene]

I think there's no generally agreed upon convention among master players on BGA. You can only guess whether you're in the majority or not with a specific aspect. And your perception might be clouded from playing a majority of games within your own bubble of friends / active hours.

Conventions have shifted in the past and if you have a good point and good and respectful conversation skills, it's possible to help it shift.

I'm 100% sure that nobody can be 100% sure about which way gives better results . Maybe you should do a tournament - but also that won't give certainty if you just play a couple of games.

The only universal truth in Hanabi seems to be that people tend to be overly convinced their opinion is the universal truth

About the subject: I'd love to adapt to allowing (number?) clues to tag nice cards when forced. So far my bubble has more players that do not allow it and prefer to clue empty (and I think that is ok too). I'd like to see the change though - just these days I mostly play to relax and avoid too much discussion, so I won't be an advocate to the matter.

[BusError]

The H-Group community agrees with smallman here. I'm ambivalent, generally, but find most people I play with prefer not to allow valid stalls.

I do think smallman should discuss conventions as trade-offs and risks, and not as 'logic'. There isn't a holy grail of logic that says color should mean play or number should mean save on chop, it's just crap people made up and agreed to.

So, what are the trade-offs?

8-Clue Stall:
Pros:
* We can mark more value in hands (e.g., saving a non-critical 2 when the following chop cards are trash).

Cons:
* We cannot mark a finesse, bluff, layer, etc. when we are at 8 clues. This includes the first turn of the game, every game.

I find the con here pretty problematic, and I think the way H-Group deals with it is by adding caveats. I don't like caveats, generally.

[UmpireNate]

Bus makes a great point; we literally can not give a finesse or bluff clue with first turn of the game with this convention .

So the convention and logic becomes “midgame” with 8 tokens we ignore the clue ?

I could get on board with 8 tokens and a number on chop is ignored; even if not critical, ie, save a valuable card , but even that has a trade off.

Generally when a number is chop saved the person receiving the clue can narrow that card down to what is critical and what is already playable, ex, “that must be b3 (playable now) or r3 (needs saved), so later I won’t need reclued to play it.

If we save arbitrary cards it’s going to need reclued later, so did we waste a token?

Idk, small man was pretty snarky to me, as my point was I’m fine with it, but it’s not hard to just state that in table notes
And then there is no issue.

[UmpireNate]

Should this logic be extended to players who are locked (5 marked cards and can’t play)? Should every clue they give be ignored too? Bc oftentimes these players have to give empty clues since they can’t play or discard .

[Jellby]

Don't forget the "free clue" flams. Those should be ignored too.

[Romain672]

So for hgroup and my post the first turn is not concerned by that 8-clue stall. Even if you don't add a caveat, the golden rule is that there is exceptions and you should ignore conventions when something isn't good, and to me, even if I understand the confusion, that's part of it.

The main point to understand is that 8-clues situations will rarely happen only once. So by allowing saves at 8 clues, it's only bad if that situation arise only once. If it arise twice, it will be neutral since you will be able to save cards during one turn, and will be forced to do a 1for1 in another turn. If it arise thrice or more, then it's better to play with 8-clue saves.
That contradict what I said earlier about wanting to not optimise hard game to get a cleaner game when things are good, like when you only get to 8 clues once.
I have no idea though how well the game with three 8 clues situations goes with hgroup.

If we save arbitrary cards it’s going to need reclued later, so did we waste a token?

In a normal game you can finesse it or wait for all the cards under it to be played before playing it. And in those kind of games with 8-clues situations, you usually finish by having a lot of playable cards available on the board, and are able to play something good anyway (something clueing an r4 tell you you don't have r4 per example, and that's frequent since you saved lots of cards on the board, and so one play can lead to 2 or 3 differents players having some plays).
Though, that's still a downside.

.

And to end, here is a summary of hgroup: https://hanabi.github.io/level-9/

Severity 0: nothing is going on, every clue will got normal meaning

Severity 1&2 doesn't apply for bga.

Severity 3: locked hand. Those are the situations where someone got a hand full of stuff, and can't discard anymore. On those spots you usually got one person giving most clue, and others just discarding and playing. If you can give a normal play clue, a normal save clue, or a tempo clue (reclueing a card to make it play) then you must give one. If you can't, you will be able to choose between: add information to already saved cards, or save any card on chop.

Severity 4: 8 clue save. Same than before, except that if you can't give something normal, you can too save any card not in chop 1. It's pretty interesting that they didn't goes for fusionning locked hand and 8-clues situations, that really show that saving random stuff and removing possibilities of finesses or bluffs have a still a good cost.

So here is what I would suggest for small groups which want to try something similar than hgroup:
- when a player is locked (no finesse position and no play available): if there is a normal play clue, a normal save clue, or if there is a clean clue to make a saved card not known playable playable to give, they must give one of those clues.
Else, any number clue which touch the chop is a save clue. (color clues are a play clue)
- when a player has 8 clues and no play available: if there is a normal play clue, a normal save clue, or if there is a clean clue to make a saved card not known playable playable to give, they must give one of those clues.
Else, if the clue touch the slot 1, it is a play clue (there can be some exception there: use context and logic).
Else, any number clue is a save clue. (color clues are undefined but usually a play clue)

Note the word 'must'. You can do deductions if someone don't respect that 'must'. And sometimes blindplay.

[BusError]

Severity 1&2 doesn't apply for bga.

I like that you're here defending L9 conventions from HGroup but immediately dismiss their single best contribution: Double Discard Avoidance. Here on BGA we have to make ambiguous double-saves and argue about triggering finesse or not because we refuse to adopt this purely superior strategy.

[Travis Hall]

I think the way H-Group deals with it is by adding caveats. I don't like caveats, generally.

Yes, this is pretty much the case. H-Group’s approach is a very detailed convention with a lot of special cases to improve the team’s chances of maxing their score, including many special cases that have a quite small incremental effect but the team who wrote the convention believe does actually improve their chances (but I don’t know if they actually did analysis of game data to see if they each worked).

And the issue with special cases is that they often only work if everybody at the table knows them. Put a player using level 9 stalls at a table with someone who does not use H-Group conventions, and some of the stalls just produce misfires. These conventions rely on pretty explicit pre game communication, and that’s fine… if you want to engage in explicit pre-game communication. (And H-Groups makes it easy to perform that communication, with their levels and formally-defined names for techniques.)

[sodiumdebt]

Hello!

My community has been asked to weigh in; I will preface with some context: I have been playing Hanabi for four years on the client the H group uses. I have tried out most conventions people on the site use (H group, Hat Guessing, Sieve, Referential Sieve, Turbo, FTL, Reactor, etc). These days I mostly play a system called Reactor 2, which was carefully developed by myself and other Hanabi Central community members to help us get full scores as often as possible. My team has played 41 designated "winstreak attempt" games so far, earning 25/25 on 40 games and 24/25 on the other. Though my understanding of BGA conventions is incomplete/inaccurate, I have significant experience developing and evaluating conventions.

It seems like you have a few differing priorities here.

(1) Simplicity. Simpler rules help team sync, and this is particularly important when you have lots of noisy communication, which BGA has: you don't discuss anything before the game, your review is heavily limited, and you'll play with people you've never seen before.
(2) Average Clue Strength. In order to win, you want to be giving powerful clues often and weak clues rarely. When disallowing stalls, you lose clue strength at the times where the best clue would be a stall and gain clue strength at the times when you want to give a finesse that would otherwise be interpreted as a stall. These situations somewhat mitigate each other and it may not be so easy to all agree on which impacts average clue strength. FWIW I suspect that, at least after the first turn at 8 clues, the average clue is somewhat stronger when allowing for nonempty stalls at the cost of (at least some) finesses.
(3) Bad Deck Mitigation. Once you are winning often enough, it starts to be worth considering optimizing some tools specifically for hard decks. The extent to which you do this depends heavily on what realistic win% you are hoping to achieve. Difficult games occur for a few (nonexclusive, non-exhaustive) ways:

• Poor Strategic Decisions / Desync. This can happen in any deck order. If this is the biggest factor contributing to your loss rate, you may want to consider finding ways to mitigate it. The main suggestions I have for this is to increase the average power of your turns or to simplify your system, i.e., care about (1) and (2).
• Dry Decks. Decks that are dry (i.e., they lack many early plays) can leave you at 8 clues while drawing, forcing you to give low value clues now, leading to issues later when your clues have to be more valuable to get all the cards played.
• Late Important Cards. Any time you are waiting for important cards near the end of the deck, you can run out of time to win. The most well-known form of this is the Bottom Deck Risk, where if a discarded 2/3/4 has its second copy on the bottom of the deck, the team cannot get full points (unless using a "Bottom Deck Blind Plays" setting). There are quite a few other concerns as well -- sometimes you need the right distribution of cards held between players, or sometimes you need a lot of clues to survive the deck order.

Late important cards tend to be a bigger problem in drier decks since you more often will have to discard good cards to cycle, and you won't have as high a clue count once you get to the end of the deck. To mitigate dry deck issues is to make sure you've configured your conventions to allow higher value clues while those decks are dry, which suggests nonempty stalls could be pretty good.

If simplicity were not a concern, I would do what Romain suggested:

Often the first time at 8 clues, it can be ok to give an empty clue to minimize the risks and make sure everyone know there is nothing good to clue on that whole board. (and make everyone see we got an 8 clue situation)
The second time you goes to 8 clues, if you clue a card which was just drawn, assuming it is playable still seem good: you want immediate tempo, and not waste more clues.
Now if it's not the case (you clue a card in chop or something else and which is valuable like a 2 or a 3), then everyone logically should get that this card wasn't playable two turns ago and still isn't playable. And everyone logically should not blind play their finesse position (since it wasn't playable two turns ago***), and should just assume a save.

This configuration should be doable to sync on and makes sure you're actually in a dry deck before allowing nonempty stalls. It should be more winning than expecting to always empty stall, so long as nobody bombs. That said, the noisy nature of BGA Hanabi communication might mean bombs are too common or that the short-term learning curve is not worth any long-term gain.

One priority I omitted is the question of "what is most intuitive?" Focusing on intuition as an objective quality can lead to somewhat rage-inducing conversations, because intuition is subjective and malleable. Many different conventions can become intuitive if you play with them enough, and we naturally have a bias to initially find our own ideas more intuitive than the ideas of others. My suggestion is to avoid strong claims based on intuition and instead focus concerns on explicitly practical problems, for example the experience of new players, the difficulty of sync, the enjoyment of the community, or expected winrate.

[UmpireNate]

Another notion is the play BEFORE 8 token situation . Discard mgmt seems to be a notion lost on BGA.

Ie, if the ones are down but 2s are slow to turn up, why are we dumping 3s and 4s when we can clue 1s to discard? I’ve clued across (skip) trash ones (known trash) and people bomb bc it’s a trash bluff, except it’s not bc there’s no bluff.

Losing those 3s and 4s waiting for 2s leads to no-play situations; hence 8 tokens

Also, more unconventional, but one bomb is nothing, it’s a resource. It partner on left has dead worthless one on finesse and I can save two or three 2s by bluff , it’s worth the bomb.

Some things to consider to lessen the amount of times we’re stuck wasting turns with 8’tokens