# Usefulness of Splendor rare moves

I'm writing a AI for the game of Splendor. While the following moves are allowed by the rules, I'm more curious if there is ever a use case to actually play them:

1. If you are buying a card, is there a benefit to use the Joker when you have the currency to afford it without the Joker?
2. If you can afford to buy a card, is there a reason to rather reserve the card?
• A generative adversarial network could allow your AI to work out the answer by playing tons of games against itself to quickly improve. Commented May 17, 2022 at 14:31
• sometimes you might want to take a card off the table simply to keep it out of somebody else's hands Commented May 17, 2022 at 16:43

A possible scenario in which you would want to reserve a card rather than buy it:

• You have:
• 10 VPs
• 3 red development cards
• 3 green development cards
• 4 black gems
• 12 VPs
• 3 white development cards
• 4 blue development cards
• 3 black development cards
• A noble is present which requires 4 white and 4 blue development cards to visit
• The only white development card visible on the board costs 3 black and gives 0 VP
• A development card is present which costs 5 black, 3 red and 3 green and gives 5 VP

You cannot purchase the 5 VP development card this turn, so you must block your opponent from taking the 0 VP development card, visiting the noble and winning. If you buy the 0 VP development card, this sets you back significantly in being able to purchase the 5 VP development card without really developing your position at all. If you instead reserve the 0 VP development card, the gold token you get means you threaten purchasing the 5 VP card on your next turn and winning yourself.

(I haven't actually checked if those specific development cards mentioned actually exist but they're pretty close to real cards, and in any case it will be easy to modify this example to cards which do exist)

• I think this is a valid scenario. I don't think it needs quite as many conditions for reserving a card you could buy to be the optimal move. Reserving a card simultaneously provides you with a joker which you could use to buy the card you need to win next turn, and prevents the opponent from buying it. So it is a much more common scenario than this answer would imply. Commented May 16, 2022 at 18:39
• @qdread I don't disagree; I just wanted to set up a situation where it was unambiguously the right thing to do rather than just probably the right thing to do. Commented May 16, 2022 at 18:51
• So in brief, you might reserve a card just for the gold token and/or to block it from the opponent, not just because you want to play it. Even without blocking, you might be one token short from playing a winning card (from your hand, hopefully), while there are none of that color available to pick in the usual way. Commented May 16, 2022 at 20:29
• @ilkkachu Yeah, that second case is probably easier to set up - you need (say) 1 white gem to be able to win but there are no white gems available, so just reserve a card for the gold token instead. Commented May 16, 2022 at 20:46

In games in general, any time there are more than two players, for any action, no matter how apparently self-defeating, it's usually possible to construct a scenario in which it is optimal. Your action might help another player, who then can block another player who is more of threat. Finding a reason for the move becomes even more of a possibility if you're engaging in collusion/trade with other players.

Even in a two-player game, there are edge cases where the scenarios you mention are optimal, especially your not buying a card case. If the reason you want the card is more to block the other player, rather than wanting the card for yourself, then reserving the card rather than buying it allows you to both block your opponent and save your gems for buying something that you actually do want.

• Great point about helping another player as a means to block a third player who is more of a threat. That sort of logic works for even more contrived scenarios than the original question. Such as spending tokens on a card instead of using the bonus discount. This could restore an empty pile allowing another play to draw tokens and then do something on a subsequent turn that blocks a third player. A very contrived, long play but theoretically possible nonetheless. Commented Oct 6, 2022 at 3:35

Short answer: removing those possible actions, will probably not reduce the win-rate too much.

Second thought: explore those actions only close to the game-end, since the chance that this action will be optimal is higher and the size of the search-space is smaller.

Responsible answer: test it, let your bot play vs bot* where bot* do not consider those moves. If the win-rate of bot* is no less than 50%, then it is safe to remove those options.

From theoretical point of view:

• action #1 can be used in order to block an opponent: for example, assume a single color stuck is depleted, and the opponent needs that color + 2 other in order to buy something. By using the joker you block that action from the opponent.
• action #2 sound to me a waste of a move, but might be optimal in a 2 players game, reserving a card keeps the options open. For example, assume 2 players game, our agent can buy one of 3 cards A,B,C while the opponent can buy A and, if takes {b,c} tokens can buy B or C respectively (depend on which tokens she collected). Here the agent's optimal strategy might be to reserve `A`, and on the next turn, buy either `B` or `C`, depends on which card the opponent made the preparations to buy.
• I disagree that these moves can be safely removed. The justification given by @Cohensius for action 1 is valid, but to me there is a more likely justification for action 1: if returning a colored token to the stack would result in 4 or more tokens being in the stack, allowing the opponent to take 2 of that color at once (which they might need to buy a key card next turn). So by using the joker instead of the colored token, you prevent the opponent from doing that. Commented May 16, 2022 at 18:41
• You are assuming that OP will have programmed his bot to properly take advantage of these moves. It may be that the moves are valuable but the OP does not know how. That is the entire purpose of their question. Commented May 16, 2022 at 18:46
• Possibly but the original question just asks whether there are "use cases" for either of the listed moves, and I think they aren't really that rare or esoteric of cases. In my experience, they don't occur every game, but probably every ten games or so. Commented May 17, 2022 at 0:48
• @qdread, you right, so I edit my answer. Commented May 17, 2022 at 6:47