Trump card: draw from the stack, no trump for wizards and jesters

- Deal the hands from the shuffled stack and draw the trump card from it,
  so the trump card is in nobody's hand and cannot be played
- A wizard or jester as trump card means no trump suit (previously a
  wizard let the dealer choose); remove the trump selection phase, the
  ChooseTrump action, the trump picker UI and the AI trump actions
- Completed tricks carry their round's trump card, so the table keeps
  showing the matching trump card while the last trick of a round is
  displayed
- Card corners are no longer rotated, so 6 and 9 cannot be confused
- Trump label explains why there is no trump (wizard, jester, last round)
- Tests: trump card never dealt, special cards give no trump, tricks keep
  their trump card

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

wizard_ai/README.md

@@ -6,7 +6,7 @@ provides everything around it.
 - `agents/` — `Agent` interface (`act(observation) -> action`) plus `RandomAgent`,
   `HeuristicAgent` (baseline / fallback) and `ModelAgent` (wraps a `Policy`)
 - `model/encoding.py` — observation → fixed-size feature vector (`FEATURE_SIZE`), and the fixed
-  action space (21 bids + 60 cards + 4 trump suits) with legal-action masks; works for 3–6 players
+  action space (21 bids + 60 cards) with legal-action masks; works for 3–6 players
 - `model/policy.py` — **the place to plug in the model**: subclass `Policy`, implement
   `select`, `update`, `save_weights`/`load`, register with `@register_policy`.
   `UniformPolicy` is a placeholder so the pipeline runs end to end.

wizard_ai/wizard_ai/agents/heuristic_agent.py

@@ -6,9 +6,7 @@ as the fallback in production while no trained model is available.
 
 from __future__ import annotations
 
-from collections import Counter
-
-from wizard_game import SUITS, Action, Bid, Card, ChooseTrump, Observation, Phase, PlayCard, Suit
+from wizard_game import Action, Bid, Card, Observation, Phase, PlayCard, Suit
 from wizard_game.rules import trick_winner
 
 from .base import Agent
@@ -18,21 +16,11 @@ class HeuristicAgent(Agent):
     name = "heuristic"
 
     def act(self, observation: Observation) -> Action:
-        if observation.phase is Phase.TRUMP_SELECTION:
-            return ChooseTrump(self._best_trump(observation.hand))
         if observation.phase is Phase.BIDDING:
             return self._bid(observation)
         return self._play(observation)
 
-    # --- Trump & bidding ----------------------------------------------------
-
-    @staticmethod
-    def _best_trump(hand: tuple[Card, ...]) -> Suit:
-        strength = Counter({suit: 0.0 for suit in SUITS})
-        for card in hand:
-            if card.is_number and card.suit is not None:
-                strength[card.suit] += 1 + card.rank / 13
-        return strength.most_common(1)[0][0]
+    # --- Bidding ------------------------------------------------------------
 
     @staticmethod
     def _expected_tricks(observation: Observation) -> float:

wizard_ai/wizard_ai/model/encoding.py

@@ -17,7 +17,6 @@ from wizard_game import (
     SUITS,
     Action,
     Bid,
-    ChooseTrump,
     Observation,
     Phase,
     PlayCard,
@@ -32,12 +31,11 @@ SCORE_SCALE = 100.0
 
 
 class ActionSpace:
-    """Fixed action indexing: bids 0-20, then the 60 cards, then 4 trump suits."""
+    """Fixed action indexing: bids 0-20, then the 60 cards."""
 
     BID_OFFSET = 0
     CARD_OFFSET = BID_OFFSET + MAX_BID + 1
-    TRUMP_OFFSET = CARD_OFFSET + DECK_SIZE
-    SIZE = TRUMP_OFFSET + len(SUITS)
+    SIZE = CARD_OFFSET + DECK_SIZE
 
     @classmethod
     def index(cls, action: Action) -> int:
@@ -45,16 +43,12 @@ class ActionSpace:
             return cls.BID_OFFSET + action.value
         if isinstance(action, PlayCard):
             return cls.CARD_OFFSET + action.card.id
-        if isinstance(action, ChooseTrump):
-            return cls.TRUMP_OFFSET + action.suit.index
         raise TypeError(f"Unknown action {action!r}")
 
     @classmethod
     def action(cls, index: int) -> Action:
         if not 0 <= index < cls.SIZE:
             raise ValueError(f"Action index out of range: {index}")
-        if index >= cls.TRUMP_OFFSET:
-            return ChooseTrump(SUITS[index - cls.TRUMP_OFFSET])
         if index >= cls.CARD_OFFSET:
             return PlayCard(card_by_id(index - cls.CARD_OFFSET))
         return Bid(index - cls.BID_OFFSET)
@@ -86,7 +80,7 @@ def _cards(cards) -> list[float]:
     return vector
 
 
-PHASES = (Phase.TRUMP_SELECTION, Phase.BIDDING, Phase.PLAYING)
+PHASES = (Phase.BIDDING, Phase.PLAYING)
 
 
 def encode_observation(obs: Observation) -> list[float]: