DominoZnif – Game Principles & AI Logic

1. Game Principles

DominoZnif is the Israeli / Eastern-Mediterranean team variant of All Fives (Muggins). Four players, two teams of two, one goal: be the first team to reach 51 points.

Setup

4 players seated in fixed positions: Bottom (you), Left, Top, Right.
Teams: Team A = Bottom & Top (partners across the table). Team B = Left & Right.
Standard double-six set: 28 tiles, pip values 0–6.
All 28 tiles are dealt — 7 per player. There is no boneyard and no drawing.

Starting a Round

Round 1: The holder of the [6|6] double opens automatically by placing it as the spinner.
Later rounds: The opener rotates clockwise. Any tile may be played first; the first double placed becomes the spinner.

      The Spinner (Znif)

      The spinner sits in the centre and can be built in all four directions (Left, Right, Up, Down), forming a cross. However, Up and Down arms are locked until both Left and Right each have at least one tile.

Play

Turns go clockwise.
On your turn, place a tile that matches one of the open arm ends. A matching double is placed crosswise.
If you cannot play, you pass ("knock").
A round ends when a player plays their last tile, or all four players pass in succession (block).

Scoring Per Move

After each tile is placed, sum the pip values at all open arm ends. If the total is a multiple of 5, the playing team scores total ÷ 5 points.

Special Spinner Rule
The central spinner counts its full double value in the open-end sum only while at least one of its Left or Right arms is still empty. Once both L and R have tiles, the spinner itself stops contributing — but other doubles on arm ends still count at full value.

Example — spinner [4|4]: Just placed, no arms → sum = 8 → not multiple of 5 → 0 pts After Left tile ending in 3, Right empty → 8+3 = 11 → 0 pts After Left 3, Right 4, Up 2, Down 1 → 3+4+2+1 = 10 (spinner no longer counts) → 2 pts

Round End

Closure — one player plays their last tile: winning team scores the loser's remaining pip sum, rounded to nearest 5.
Block — all four pass: lower-pip team wins that rounded sum. Equal pips = no points awarded.

Game End

First team to reach 51 points wins. The current round still finishes. If both teams end a round at ≥ 51 with equal scores, play continues until one team strictly leads.

2. AI Architecture

The AI in DominoZnif is not random. Every move is calculated. Here is how it thinks.

Difficulty Levels & Rollouts

The AI has 9 difficulty levels. The higher the level, the more simulations it runs before choosing a move:

Difficulty	Simulations (N)	Method
Novice	0	Single-ply heuristic only
Beginner	200	Heuristic-guided rollouts
Casual	350	Heuristic-guided rollouts
Average	500	Heuristic-guided rollouts
Skilled	650	Heuristic-guided rollouts
Sharp	800	Heuristic-guided rollouts
Expert	950	Heuristic-guided rollouts
Master	1,100	Heuristic-guided rollouts
Legend	1,250	Heuristic-guided rollouts

How Rollouts Work

For each possible move, the AI runs N simulations:

Clones the board and all players onto a scratch copy.
Randomly samples the opponents' unknown tiles (based only on what a real player could observe).
Plays the round to completion using the heuristic for all seats.
Records the score difference: my team − opponent team.
Averages the result over N simulations.

The move with the best average wins.

What the AI Is Allowed to See

The AI plays fair. It can only use information a real human player could observe:

Its own tiles.
Tiles placed on the board.
Pass events and scoring events at the table.

It never peeks at opponents' hands. Unknown tiles are sampled randomly from a pool that excludes placed tiles and the AI's own hand.

3. How the Heuristic Evaluates a Move

For Novice difficulty, and as the evaluator inside every rollout simulation, the AI scores each legal move using a weighted sum of these factors:

Factor	What it does
ScoreVal	Rewards scoring points right now.
PrvOpp	Penalises leaving a big scoring opportunity for the next opponent.
BlkVal	Rewards forcing the opponent to pass.
HvyDump	Encourages playing heavy doubles early — they hurt if you're stuck with them at round end.
SelfDom	Rewards keeping your dominant pip value playable on the board.
TeamDom	Supports your partner's likely future plays based on inference.
ArmAln	Leaves open arm ends showing values you dominate.
Kill5	Closes a 5-end before the opponent can score from it. (+60)
SelfDbl	Soft bonus for denying the opponent their strongest double reply. (+50)
BlkChain	Hard bonus when a move forces all three opponents to pass AND your team has lighter tiles. (+500)
NoReply	Bonus when you score AND the opponent cannot reply: +100 if net gain ≥ 3, +200 if ≥ 5.

Endgame Awareness

When a team's score approaches 51, the AI adjusts: if a move scores points, it ignores the opponent-reply penalty and takes the points. Winning is more important than defensive positioning when the game is close.

4. Opening Move Shortcut

Before running any simulation, the AI checks for a dominant opening:

The player is the round opener and the board is empty.
The player holds [X|X] AND at least 4 other tiles of pip X.
X ≤ 5 (at X = 6, an opponent could score immediately from the spinner sum).

If all conditions are met, the AI immediately opens with [X|X] as the spinner — no simulation needed. This is the strongest possible opening position.