12-Angry-Agent / core /orchestrator.py
Blu3Orange
feat: Introduce argument direction handling and enhance conviction mechanics for juror interactions
373ff24
"""Orchestrator for 12 Angry Agents deliberation.
Handles turn management, speaker selection, and deliberation flow.
"""
import random
from dataclasses import dataclass, field
from typing import TYPE_CHECKING
from core.game_state import GameState, DeliberationTurn, GamePhase, ArgumentDirection
from core.models import JurorConfig, JurorMemory
from core.conviction import calculate_conviction_change, apply_conviction_update
if TYPE_CHECKING:
from agents.smolagent_juror import SmolagentJuror
from case_db.models import CriminalCase
@dataclass
class SpeakerWeight:
"""Weight information for speaker selection."""
juror_id: str
weight: float
reason: str
@dataclass
class TurnResult:
"""Result of a single turn in deliberation."""
turn: DeliberationTurn
conviction_changes: dict[str, float]
vote_changes: list[tuple[str, str, str]]
reasoning_steps: list[str] = field(default_factory=list)
tool_calls: list[str] = field(default_factory=list)
class TurnManager:
"""Manages fair speaker selection with weighted queue.
Selection priorities:
1. Jurors "on the fence" (conviction 0.35-0.65) - most interesting
2. Jurors who haven't spoken recently - fairness
3. Jurors with high influence - they drive conversation
4. Some randomness to keep things unpredictable
"""
ON_FENCE_BONUS = 2.0
RECENCY_PENALTY = 0.3
INFLUENCE_WEIGHT = 1.5
RANDOM_FACTOR = 0.3
RECENCY_WINDOW = 2
def __init__(self):
self.speaker_history: list[list[str]] = []
def select_speakers(
self,
game_state: GameState,
juror_configs: list[JurorConfig],
juror_memories: dict[str, JurorMemory],
num_speakers: int = None,
exclude_player: bool = True
) -> list[str]:
"""Select speakers for the next round using weighted selection."""
if num_speakers is None:
num_speakers = random.randint(1, 3)
eligible = [
c for c in juror_configs
if not (exclude_player and c.is_player())
]
if not eligible:
return []
weights = self._calculate_weights(
eligible,
juror_memories,
game_state.round_number
)
selected = self._weighted_select(weights, min(num_speakers, len(eligible)))
self.speaker_history.append(selected)
game_state.speaking_queue = selected
return selected
def _calculate_weights(
self,
configs: list[JurorConfig],
memories: dict[str, JurorMemory],
current_round: int
) -> list[SpeakerWeight]:
"""Calculate selection weight for each juror."""
weights = []
for config in configs:
jid = config.juror_id
memory = memories.get(jid)
base_weight = 0.5 + (config.influence * self.INFLUENCE_WEIGHT)
if memory:
conviction = memory.current_conviction
fence_distance = abs(conviction - 0.5)
if fence_distance < 0.15:
fence_bonus = self.ON_FENCE_BONUS * (1 - fence_distance / 0.15)
else:
fence_bonus = 0.0
else:
fence_bonus = 0.0
recency_multiplier = 1.0
reason_parts = []
for rounds_ago, speakers in enumerate(reversed(self.speaker_history[-self.RECENCY_WINDOW:])):
if jid in speakers:
penalty = self.RECENCY_PENALTY ** (rounds_ago + 1)
recency_multiplier *= penalty
reason_parts.append(f"spoke {rounds_ago + 1} rounds ago")
break
volatility_bonus = config.volatility * 0.5
weight = (base_weight + fence_bonus + volatility_bonus) * recency_multiplier
weight += random.uniform(0, self.RANDOM_FACTOR)
reasons = []
if fence_bonus > 0:
reasons.append(f"on fence (+{fence_bonus:.2f})")
if recency_multiplier < 1.0:
reasons.append(f"recent speaker (x{recency_multiplier:.2f})")
if config.influence > 0.6:
reasons.append("high influence")
if config.volatility > 0.6:
reasons.append("volatile")
weights.append(SpeakerWeight(
juror_id=jid,
weight=max(0.1, weight),
reason=", ".join(reasons) if reasons else "baseline"
))
return weights
def _weighted_select(
self,
weights: list[SpeakerWeight],
count: int
) -> list[str]:
"""Select jurors using weighted random selection without replacement."""
selected = []
remaining = list(weights)
for _ in range(count):
if not remaining:
break
total = sum(w.weight for w in remaining)
if total <= 0:
break
r = random.uniform(0, total)
cumulative = 0
for i, w in enumerate(remaining):
cumulative += w.weight
if r <= cumulative:
selected.append(w.juror_id)
remaining.pop(i)
break
return selected
def reset(self):
"""Reset speaker history for new game."""
self.speaker_history = []
class OrchestratorAgent:
"""Master agent that coordinates the deliberation.
Handles:
- Game phase transitions
- Turn management and speaker selection
- Processing arguments and reactions
- Vote tracking and stability detection
"""
def __init__(
self,
juror_configs: list[JurorConfig],
juror_agents: dict[str, "SmolagentJuror"],
case: "CriminalCase"
):
self.juror_configs = juror_configs
self.juror_agents = juror_agents
self.case = case
self.turn_manager = TurnManager()
self.state = GameState(case_id=case.case_id)
for jid, agent in juror_agents.items():
self.state.votes[jid] = agent.get_vote()
self.state.conviction_scores[jid] = agent.memory.current_conviction
@property
def game_state(self) -> GameState:
"""Get current game state."""
return self.state
def get_juror_memories(self) -> dict[str, JurorMemory]:
"""Get memory state for all jurors."""
return {jid: agent.memory for jid, agent in self.juror_agents.items()}
async def run_deliberation_round(
self,
num_speakers: int = None
) -> list[TurnResult]:
"""Run a single round of deliberation."""
self.state.round_number += 1
results = []
votes_at_start = dict(self.state.votes)
speakers = self.turn_manager.select_speakers(
self.state,
self.juror_configs,
self.get_juror_memories(),
num_speakers=num_speakers,
exclude_player=True
)
for speaker_id in speakers:
result = await self._process_speaker_turn(speaker_id)
if result:
results.append(result)
if self.state.votes == votes_at_start:
self.state.rounds_without_change += 1
else:
self.state.rounds_without_change = 0
return results
async def _process_speaker_turn(self, speaker_id: str) -> TurnResult | None:
"""Process a single speaker's turn."""
agent = self.juror_agents.get(speaker_id)
if not agent:
return None
try:
# Generate argument - SmolagentJuror always returns (turn, reasoning_steps)
turn, reasoning_data = await agent.generate_argument(self.case, self.state)
# Extract reasoning steps for UI
reasoning_steps = []
if reasoning_data:
reasoning_steps = [
f"Step {s.step_number}: {s.action} - {s.content[:100]}"
if hasattr(s, 'step_number') else str(s)
for s in reasoning_data
]
# Extract tool calls
tool_calls = agent.last_tool_calls if hasattr(agent, 'last_tool_calls') else []
# Select active listeners for full processing
active_listeners = self._select_active_listeners(turn)
# Process reactions from other jurors
conviction_changes = {}
vote_changes = []
for other_id, other_agent in self.juror_agents.items():
if other_id == speaker_id:
continue
old_vote = self.state.votes.get(other_id)
# Calculate base strength based on speaker influence
base_strength = 0.08 + (agent.config.influence * 0.07)
if other_id in active_listeners:
base_strength *= 1.2
# Use direction-aware conviction calculation
delta = calculate_conviction_change(
other_agent.config,
other_agent.memory,
turn, # turn now includes direction
base_strength=base_strength
)
# Store argument in memory (don't apply delta here - do it via apply_conviction_update)
other_agent.receive_argument(turn, 0.0)
conviction_changes[other_id] = delta
turn.impact[other_id] = delta
# Apply conviction update with hysteresis
vote_flipped, new_vote = apply_conviction_update(other_agent.memory, delta)
if vote_flipped and new_vote:
self.state.votes[other_id] = new_vote
vote_changes.append((other_id, old_vote, new_vote))
self.state.conviction_scores[other_id] = other_agent.memory.current_conviction
self.state.deliberation_log.append(turn)
return TurnResult(
turn=turn,
conviction_changes=conviction_changes,
vote_changes=vote_changes,
reasoning_steps=reasoning_steps,
tool_calls=tool_calls,
)
except Exception as e:
print(f"Error processing turn for {speaker_id}: {e}")
return None
def _select_active_listeners(
self,
turn: DeliberationTurn,
max_active: int = 3
) -> list[str]:
"""Select jurors for full agent processing (active listeners)."""
active = []
for jid, agent in self.juror_agents.items():
if jid == turn.speaker_id:
continue
if 0.35 < agent.memory.current_conviction < 0.65:
active.append((jid, 3))
elif agent.config.influence > 0.7:
active.append((jid, 2))
elif turn.target_id == jid:
active.append((jid, 3))
elif len(agent.memory.conviction_history) > 1:
recent_change = abs(
agent.memory.conviction_history[-1] -
agent.memory.conviction_history[-2]
) if len(agent.memory.conviction_history) >= 2 else 0
if recent_change > 0.1:
active.append((jid, 2))
else:
active.append((jid, 1))
active.sort(key=lambda x: x[1], reverse=True)
return [jid for jid, _ in active[:max_active]]
def process_player_argument(
self,
content: str,
argument_type: str,
target_id: str | None = None
) -> TurnResult:
"""Process an argument from the human player."""
# Determine direction from player's chosen side
direction = (
ArgumentDirection.PROSECUTION
if self.state.player_side == "prosecute"
else ArgumentDirection.DEFENSE
)
turn = DeliberationTurn(
round_number=self.state.round_number,
speaker_id="juror_7",
speaker_name="You",
argument_type=argument_type,
direction=direction,
content=content,
target_id=target_id
)
conviction_changes = {}
vote_changes = []
for juror_id, agent in self.juror_agents.items():
old_vote = self.state.votes.get(juror_id)
# Calculate base strength (player has moderate influence)
base_strength = 0.10
if target_id == juror_id:
base_strength *= 1.5
delta = calculate_conviction_change(
agent.config,
agent.memory,
turn,
base_strength=base_strength
)
# Store argument in memory (don't apply delta here)
agent.receive_argument(turn, 0.0)
conviction_changes[juror_id] = delta
turn.impact[juror_id] = delta
# Apply conviction update with hysteresis
vote_flipped, new_vote = apply_conviction_update(agent.memory, delta)
if vote_flipped and new_vote:
self.state.votes[juror_id] = new_vote
vote_changes.append((juror_id, old_vote, new_vote))
self.state.conviction_scores[juror_id] = agent.memory.current_conviction
self.state.deliberation_log.append(turn)
return TurnResult(
turn=turn,
conviction_changes=conviction_changes,
vote_changes=vote_changes
)
def process_external_argument(
self,
speaker_id: str,
speaker_name: str,
content: str,
direction: ArgumentDirection, # REQUIRED - no fallback
argument_type: str = "logical",
target_id: str | None = None
) -> TurnResult:
"""Process an argument from an external MCP agent.
Similar to process_player_argument but with configurable speaker identity.
Used when external AI agents participate via MCP.
Args:
speaker_id: Juror seat ID (e.g., "juror_3")
speaker_name: Display name for the speaker
content: Argument text
direction: REQUIRED - "prosecution", "defense", or "neutral"
argument_type: Type of argument (default: "logical")
target_id: Optional juror_id to address directly
Returns:
TurnResult with conviction changes and vote changes
"""
turn = DeliberationTurn(
round_number=self.state.round_number,
speaker_id=speaker_id,
speaker_name=speaker_name,
argument_type=argument_type,
direction=direction,
content=content,
target_id=target_id
)
conviction_changes = {}
vote_changes = []
for juror_id, agent in self.juror_agents.items():
if juror_id == speaker_id:
continue
old_vote = self.state.votes.get(juror_id)
# Calculate base strength (external agents have moderate influence)
base_strength = 0.10
if target_id == juror_id:
base_strength *= 1.5
delta = calculate_conviction_change(
agent.config,
agent.memory,
turn,
base_strength=base_strength
)
# Store argument in memory (don't apply delta here)
agent.receive_argument(turn, 0.0)
conviction_changes[juror_id] = delta
turn.impact[juror_id] = delta
# Apply conviction update with hysteresis
vote_flipped, new_vote = apply_conviction_update(agent.memory, delta)
if vote_flipped and new_vote:
self.state.votes[juror_id] = new_vote
vote_changes.append((juror_id, old_vote, new_vote))
self.state.conviction_scores[juror_id] = agent.memory.current_conviction
self.state.deliberation_log.append(turn)
return TurnResult(
turn=turn,
conviction_changes=conviction_changes,
vote_changes=vote_changes
)
def set_player_side(self, side: str) -> None:
"""Set the player's chosen side."""
self.state.player_side = side
self.state.phase = GamePhase.DELIBERATION
player_vote = "guilty" if side == "prosecute" else "not_guilty"
self.state.votes["juror_7"] = player_vote
self.state.conviction_scores["juror_7"] = 0.8 if side == "prosecute" else 0.2
def check_should_end(self) -> bool:
"""Check if deliberation should end."""
return self.state.should_end_deliberation()
def get_verdict(self) -> dict:
"""Get the final verdict information."""
guilty, not_guilty = self.state.get_vote_tally()
if self.state.is_unanimous():
verdict = "GUILTY" if guilty == 12 else "NOT GUILTY"
unanimous = True
else:
verdict = "HUNG JURY"
unanimous = False
return {
"verdict": verdict,
"unanimous": unanimous,
"guilty_count": guilty,
"not_guilty_count": not_guilty,
"rounds": self.state.round_number,
"ended_by": self._get_end_reason()
}
def _get_end_reason(self) -> str:
"""Get the reason deliberation ended."""
if self.state.is_unanimous():
return "unanimous_verdict"
elif self.state.rounds_without_change >= self.state.stability_threshold:
return "votes_stabilized"
elif self.state.round_number >= self.state.max_rounds:
return "max_rounds_reached"
return "unknown"
def reset(self) -> None:
"""Reset for a new game."""
self.turn_manager.reset()
self.state = GameState(case_id=self.case.case_id if self.case else "")