Formal Specification
Three-Seven — Formal Specification
Italy · card · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: directGap-fills: 1
# Tressette — AI-optimized specification
## General idea
A four-agent trick-taking game played in fixed partnerships without trumps, using an inverted card hierarchy where Threes and Twos dominate Aces and face cards. Agents must strictly follow suit and cooperate with incomplete information to capture high-value cards (Aces = 1 pt, Threes/Twos/Faces = 1/3 pt) and the 1-point final trick out of an exact 11-point deal budget.
## Formal model
- **Abstraction taken:** None; rules form a completely discrete state machine over imperfect information.
- **Player/agent count:** 4 agents arranged as Team 0 (Agents 0, 2) vs Team 1 (Agents 1, 3).
- **Turn model:** Strictly sequential, 4-step trick rotation; winner of each trick leads the next.
## Formal rules
R1. Deck structure: 40 cards, 4 suits. Trick hierarchy within suit: .
R2. Card values: , for , for .
R3. Deal: All 40 cards are dealt (10 private cards per agent). Target match score is .
R4. Trick execution:
- Leader plays any card .
- Subsequent agents in rotation must play a card of if they hold at least one such card in hand.
- If void in , the agent may discard any card in hand.
R5. Trick resolution: The trick is won by the agent who played the highest-ranked card matching the suit led. Off-suit cards cannot win.
R6. Deal scoring:
- Points from cards: (integer division / dropping fractions).
- Last trick: point awarded to the team winning trick 10.
- Total points available per deal = 11.
R7. [gap-fill] Match win condition: The match ends at the conclusion of any deal where at least one team reaches points. If both teams reach , the team with the strictly higher total wins. If tied at , additional full deals are played until the tie is broken.
## Edge cases
- Tie at the end condition: Ties at trigger additional complete deals until one team holds a strictly higher integer score (R7).
- A player with no legal move: Impossible; hand size is non-empty for all 10 tricks, and at least one card in hand is always legal (either matching suit or discard).
- Repetition / cycles: Impossible; exactly 10 tricks are played per deal, strictly consuming all 40 cards.
- Simultaneous claims on the same resource or position: Not applicable; turns are strictly sequential.
- Player elimination mid-round; turn order repair: Not applicable; 4 fixed agents throughout.
- Resource exhaustion (pieces, tokens, space, time): The 40-card deal exhausts in exactly 10 tricks.
- Termination guarantee: Each non-trivial deal awards 11 total points across the two teams; total points monotonically increase, guaranteeing match termination in at most a few deals.
## State representation
- Persistent state:
- `score_team_0`, `score_team_1`: integers.
- `target_score`: 21.
- Per-deal state:
- `hands`: 4 lists of cards (size 10 down to 0).
- `tricks_history`: list of completed 4-card tricks.
- `current_trick`: list of cards played in current trick.
- `captured_team_0`, `captured_team_1`: lists of captured cards.
- `trick_leader`: agent ID .
- Observable information: Discarded cards on each trick, trick winners, and cumulative scores are public. Partner and opponent hands are private.
## Action space
- `play_card(card_id: int)`: Precondition: Must be of led suit if player holds cards of led suit, otherwise any card in hand.
## Playability notes
- Single-player-vs-AI: Engine-ready; excellent domain for Information-Set Monte Carlo Tree Search (IS-MCTS) and neural network belief state modeling.
- Group games (n>4): Not applicable; strictly 4-agent partnership.
- Physical-skill games: Direct discrete card play.
- What is lost in the adaptation: Traditional spoken and gestural signals (*busso*, *volo*, *striscio*) are omitted; pure card-play logic is modeled.
## Implementation hints
Branching factor per turn is small (1 to 10 legal cards, restricted heavily by the follow-suit constraint).
First milestone: Build the follow-suit validator and 11-point deal scoring accumulator before developing partnership bidding or inference models.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Three-Seven (Italy). (n.d.). Games of Countries. https://gamesofcountries.com/games/it-italy/tressette
@misc{gamesofcountries-it-italy-tressette,
title = {Three-Seven (Italy)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/it-italy/tressette}
}