Reduce LLM call rate: shorter convos, less frequent, hard cap

- max_turns: 6 → 3 (halves turns per conversation)
- initiation base_chance: /20 → /40 (halves conversation start rate)
- max simultaneous conversations: dynamic → hard cap of 2

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/soci/actions/conversation.py

@@ -35,7 +35,7 @@ class Conversation:
     turns: list[ConversationTurn] = field(default_factory=list)
     topic: str = ""
     is_active: bool = True
-    max_turns: int = 6
+    max_turns: int = 3
 
     @property
     def is_finished(self) -> bool:

src/soci/actions/social.py

@@ -17,7 +17,7 @@ def should_initiate_conversation(agent: Agent, other_id: str, clock: SimClock) -
         return False
 
     # Extraversion drives conversation initiation
-    base_chance = agent.persona.extraversion / 20.0  # 0.05 to 0.5
+    base_chance = agent.persona.extraversion / 40.0  # 0.025 to 0.25
 
     # Boost if social need is low (lonely agents seek conversation)
     if agent.needs.social < 0.3:

src/soci/engine/simulation.py

@@ -456,8 +456,8 @@ class Simulation:
 
     async def _handle_social_interactions(self, agents: list[Agent]) -> None:
         """Check if any idle co-located agents should start conversations."""
-        # Don't flood with conversations — scale with population
-        max_convos = max(3, len(self.agents) // 5)
+        # Hard cap: at most 2 simultaneous conversations to keep LLM calls low
+        max_convos = 2
         if len(self.active_conversations) >= max_convos:
             return