Fix Gemini conversations: bump LLM budget to 2 and remove double probability gate

With budget=1, action decisions consumed the entire per-tick allowance,
leaving zero budget for conversations. Also, active conversations were
gated by llm_call_probability on every turn — now they always continue
once started.

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

src/soci/api/server.py

@@ -115,9 +115,10 @@ async def simulation_loop(sim: Simulation, db: Database, tick_delay: float = 2.0
                 sim._skip_llm_this_tick = False
                 if is_rate_limited:
                     # Rate-limited providers: tight budget — probability slider does the fine-tuning.
-                    # Gemini free tier: 4 RPM, ~1500 RPD → budget=1 + prob=0.45 ≈ 150 calls/h (10h).
+                    # Gemini free tier: 4 RPM, ~1500 RPD → budget=2 + prob=0.10 ≈ 15 calls/h.
+                    # Budget=2 so 1 action + 1 conversation can co-exist per tick.
                     sim._max_convos_this_tick = 1
-                    sim._max_llm_calls_this_tick = 1
+                    sim._max_llm_calls_this_tick = 2
                 else:
                     # Ollama / Claude: soft cap to keep ticks responsive
                     sim._max_convos_this_tick = 3

src/soci/engine/simulation.py

@@ -279,7 +279,9 @@ class Simulation:
                 if next_speaker_id:
                     responder = self.agents.get(next_speaker_id[0])
                     other = self.agents.get(last_speaker) if last_speaker else None
-                    if responder and other and random.random() < self.llm_call_probability:
+                    if responder and other:
+                        # Always continue active conversations — they already passed
+                        # the probability gate when initiated; don't double-gate them.
                         conv_coros.append(
                             continue_conversation(conv, responder, other, self.llm, self.clock)
                         )