Fix think thoroughly final question (#133)

prediction_market_agent/agents/think_thoroughly_agent/prompts.py

@@ -93,8 +93,10 @@
 You should determine the probability of the SCENARIO SCENARIO_TO_ASSESS being true, 
 considering the probabilities of the other related SCENARIOs.
 
+SCENARIOS_WITH_PROBABILITIES: 
+{scenarios_with_probabilities}
+
 SCENARIO_TO_ASSESS: {scenario_to_assess}
-SCENARIOS_WITH_PROBABILITIES: {scenarios_with_probabilities}
 """
 
 PROBABILITY_CLASS_OUTPUT = """

prediction_market_agent/agents/think_thoroughly_agent/think_thoroughly_agent.py

@@ -149,7 +149,7 @@ def generate_prediction_for_one_outcome(
         inputs = {"sentence": sentence}
         if previous_scenarios_and_answers:
             inputs["previous_scenarios_with_probabilities"] = "\n".join(
-                f"- Scenario '{s}' has probability of happening {a.p_yes * 100:.2f}%, because {a.reasoning}"
+                f"- Scenario '{s}' has probability of happening {a.p_yes * 100:.2f}% with confidence {a.confidence * 100:.2f}%, because {a.reasoning}"
                 for s, a in previous_scenarios_and_answers
             )
 
@@ -171,7 +171,7 @@ def generate_prediction_for_one_outcome(
             return None
 
     def generate_final_decision(
-        self, scenarios_with_probabilities: list[t.Tuple[str, Answer]]
+        self, question: str, scenarios_with_probabilities: list[t.Tuple[str, Answer]]
     ) -> Answer:
         predictor = self._get_predictor()
 
@@ -188,13 +188,15 @@ def generate_final_decision(
             verbose=2,
         )
 
+        logger.info(f"Starting to generate final decision for '{question}'.")
         crew.kickoff(
             inputs={
-                "scenarios_with_probabilities": [
-                    (i[0], i[1].model_dump()) for i in scenarios_with_probabilities
-                ],
+                "scenarios_with_probabilities": "\n".join(
+                    f"- Scenario '{s}' has probability of happening {a.p_yes * 100:.2f}% with confidence {a.confidence * 100:.2f}%, because {a.reasoning}"
+                    for s, a in scenarios_with_probabilities
+                ),
                 "number_of_scenarios": len(scenarios_with_probabilities),
-                "scenario_to_assess": scenarios_with_probabilities[0][0],
+                "scenario_to_assess": question,
             }
         )
         output = Answer.model_validate_json(task_final_decision.output.raw_output)
@@ -233,7 +235,7 @@ def answer_binary_market(
                 )
 
         final_answer = (
-            self.generate_final_decision(scenarios_with_probs)
+            self.generate_final_decision(question, scenarios_with_probs)
             if scenarios_with_probs
             else None
         )

prediction_market_agent/tools/streamlit_utils.py

@@ -11,11 +11,7 @@ def loguru_streamlit_sink(log: "Message") -> None:
     record = log.record
     level = record["level"].name
 
-    message = record["message"]
-    # Replace escaped newlines with actual newlines.
-    message = message.replace("\\n", "\n")
-    # Fix malformed dollar signs in the messages.
-    message = message.replace("$", "\$")
+    message = streamlit_escape(record["message"])
 
     if level == "ERROR":
         st.error(message, icon="❌")
@@ -35,3 +31,15 @@ def add_sink_to_logger() -> None:
     Needs to be behind a cache decorator, so it only runs once per streamlit session (otherwise we would see duplicated messages).
     """
     logger.add(loguru_streamlit_sink)
+
+
+def streamlit_escape(message: str) -> str:
+    """
+    Escapes the string for streamlit writes.
+    """
+    # Replace escaped newlines with actual newlines.
+    message = message.replace("\\n", "\n")
+    # Fix malformed dollar signs in the messages.
+    message = message.replace("$", "\$")
+
+    return message

scripts/agent_app.py

@@ -26,7 +26,10 @@
 from prediction_market_agent.agents.think_thoroughly_agent.deploy import (
     DeployableThinkThoroughlyAgent,
 )
-from prediction_market_agent.tools.streamlit_utils import add_sink_to_logger
+from prediction_market_agent.tools.streamlit_utils import (
+    add_sink_to_logger,
+    streamlit_escape,
+)
 
 AGENTS: list[
     t.Type[DeployableKnownOutcomeAgent] | t.Type[DeployableThinkThoroughlyAgent]
@@ -113,4 +116,8 @@
                 bet_amount = agent.calculate_bet_amount(answer, market)
 
         st.warning(f"Took {costs.time / 60:.2f} minutes and {costs.cost:.2f} USD.")
-        st.success(f"Would bet {bet_amount.amount} {bet_amount.currency} on {answer}!")
+        st.success(
+            streamlit_escape(
+                f"Would bet {bet_amount.amount} {bet_amount.currency} on {answer}!"
+            )
+        )