Skip to main content
Formal Specification

Differenzler Jass (Exact point prediction Jass) — Formal Specification

Switzerland · card · Confidence: high

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: directGap-fills: 1
# Differenzler Jass — AI-optimized specification ## General idea A four-player individual imperfect-information trick-taking game. Agents inspect their private 9-card hands and commit to an exact integer point prediction Ai∈[0,157]. During play, agents navigate standard Swiss Jass suit-following and trump constraints, aiming not to maximize tricks, but to force their captured card point total Pi to match Ai exactly, minimizing the absolute deviation penalty |Ai−Pi|. ## Formal model - **Abstraction taken:** None required (`direct` fidelity). The 36-card deck, point distribution (sum =157), trick rules, and difference evaluation are completely discrete. - **Player/agent count:** 4 agents (0,1,2,3). - **Turn model:** Sequential hidden prediction phase, followed by 9 sequential tricks of 4 cards each. ## Formal rules R1. Deal and Trump Setup: 36-card deck C dealt completely (9 cards per agent). Trump suit T is publicly established. Total deck value = 157 points (including 5 points for trick 9). R2. Prediction Phase: Each agent i privately commits to an integer prediction Ai∈[0,157]. Predictions are stored secretly until hand completion. R3. Trick Play Constraints: - Leader plays any card. - Follower constraints: 1. Must follow suit led (*Farbzwang*). 2. If void in suit led, may play a trump card or discard any non-trump card. 3. Under-trumping (*Untertrumpfen*) is prohibited: a player may not play a lower trump than one already played to the trick unless holding only trumps. 4. Bauer exception: The Jack of Trump is not required to follow suit when trump is led. - Trick winner collects trick cards and leads next trick. R4. Hand Evaluation: - At the end of 9 tricks, each agent sums the card values in all collected tricks: Pi=∑c∈tricksivalue(c) (adding +5 if winning trick 9). Note that ∑i=03Pi=157. - Hand penalty: Di,h=|Ai−Pi|. R5. Match Score & Victory: - Across Hmax⁡=8 deals, each agent accumulates penalty points: TotalPenaltyi=∑h=1Hmax⁡Di,h. - The agent with the lowest total penalty score wins the tournament. ## Edge cases - **Tie at the end condition:** [gap-fill] If two agents tie with lowest cumulative penalty points after Hmax⁡ deals, the agent with the highest count of exact hits (Di,h=0) wins. - **A player with no legal move:** Standard discard rules guarantee a legal play is always available. - **Repetition / cycles:** Exactly 9 tricks of 4 cards per hand over Hmax⁡ hands; cycling is impossible. - **Simultaneous claims on the same resource or position:** Turns and card plays occur in strict sequential clockwise order. - **Player elimination mid-round; turn order repair:** No mid-round eliminations; all 4 players participate. - **Resource exhaustion:** Hand strictly ends after 9 tricks (all 36 cards played). - **Termination guarantee:** Guaranteed by finite 9-trick structure and Hmax⁡=8 hand limit. ## State representation - **Persistent state:** Cumulative penalty vector 𝐃∈ℕ4, exact-hit counter 𝐙∈ℕ4, hand counter h∈[1,8]. - **Per-hand state:** Agent hands Hi⊂C, trump suit T, secret predictions {Ai}, current trick cards, collected trick points {Pi}. - **Observable information:** Imperfect information. Each agent sees own cards, trump suit, and past trick cards. Opponent hands and opponent secret predictions Aj are private until hand completion. ## Action space - `commit_prediction(points)`: Integer A∈[0,157]. - `play_card(card)`: Play card from hand obeying R3 constraints. ## Playability notes - **Single-player-vs-AI:** Outstanding individual tactical challenge; AI agents must evaluate hand potential, estimate trick distributions, and adapt mid-hand to "bleed" unwanted points. - **Group games (n>4):** Standard 4-player game; can be played in parallel tables. - **Physical-skill games:** Not applicable (pure discrete card game). - **What is lost in the adaptation:** The tense silence during card reveals on television and pub table sighs when an unintended King is captured. ## Implementation hints - Implement the point-tracking vector Pi after each trick, with card value lookup depending on whether c∈T or c∉T. - Suggested first milestone: Implement the 9-trick play engine and exact absolute difference evaluation before designing prediction heuristics.

Download as JSON

Download the full game corpus entry for this game as structured JSON.

Download JSON
Cite this page

Differenzler Jass (Exact point prediction Jass) (Switzerland). (n.d.). Games of Countries. https://gamesofcountries.com/games/ch-switzerland/differenzler-jass

@misc{gamesofcountries-ch-switzerland-differenzler-jass,
  title = {Differenzler Jass (Exact point prediction Jass) (Switzerland)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/ch-switzerland/differenzler-jass}
}