Workflow update.

refactored single_run, eval on last question run too.

src/workflows/factory.py

@@ -15,8 +15,20 @@ Given a question clue, output your most likely guess in a couple words with a ca
 """
 
 
-def create_simple_workflow():
-    pass
+def create_empty_bonus_workflow():
+    return Workflow(
+        inputs=["leadin", "part"],
+        outputs={"answer": None, "confidence": None, "explanation": None},
+        steps={},
+    )
+
+
+def create_empty_tossup_workflow():
+    return TossupWorkflow(
+        inputs=["question_text"],
+        outputs={"answer": None, "confidence": None},
+        steps={},
+    )
 
 
 def create_first_step_input_fields() -> list[InputField]:

src/workflows/qb_agents.py

@@ -20,6 +20,7 @@ def _get_workflow_response(
 class TossupResult(TypedDict):
     answer: str
     confidence: float
+    logprob: float | None
     buzz: bool
     question_fragment: str
     position: int
@@ -62,6 +63,27 @@ class QuizBowlTossupAgent:
             if out_var not in workflow.outputs:
                 raise ValueError(f"Output variable {out_var} not found in workflow outputs")
 
+    def _single_run(self, question_run: str, position: int) -> TossupResult:
+        """Process a single question run."""
+        answer_var_step = self.workflow.outputs["answer"].split(".")[0]
+        workflow_output, response_time = _get_workflow_response(
+            self.workflow, {self.external_input_variable: question_run}, logprob_step=answer_var_step
+        )
+        final_outputs = workflow_output["final_outputs"]
+        buzz = self.workflow.buzzer.run(final_outputs["confidence"], logprob=workflow_output["logprob"])
+        result: TossupResult = {
+            "position": position,
+            "answer": final_outputs["answer"],
+            "confidence": final_outputs["confidence"],
+            "logprob": workflow_output["logprob"],
+            "buzz": buzz,
+            "question_fragment": question_run,
+            "step_contents": workflow_output["step_contents"],
+            "step_outputs": workflow_output["intermediate_outputs"],  # Include intermediate step outputs
+            "response_time": response_time,
+        }
+        return result
+
     def run(self, question_runs: list[str], early_stop: bool = True) -> Iterable[TossupResult]:
         """Process a tossup question and decide when to buzz based on confidence.
 
@@ -82,28 +104,14 @@ class QuizBowlTossupAgent:
         """
         for i, question_text in enumerate(question_runs):
             # Execute the complete workflow
-            answer_var_step = self.workflow.outputs["answer"].split(".")[0]
-            workflow_output, response_time = _get_workflow_response(
-                self.workflow, {self.external_input_variable: question_text}, logprob_step=answer_var_step
-            )
-            final_outputs = workflow_output["final_outputs"]
-            buzz = self.workflow.buzzer.run(final_outputs["confidence"], logprob=workflow_output["logprob"])
-            result: TossupResult = {
-                "position": i + 1,
-                "answer": final_outputs["answer"],
-                "confidence": final_outputs["confidence"],
-                "logprob": workflow_output["logprob"],
-                "buzz": buzz,
-                "question_fragment": question_text,
-                "step_contents": workflow_output["step_contents"],
-                "step_outputs": workflow_output["intermediate_outputs"],  # Include intermediate step outputs
-                "response_time": response_time,
-            }
+            result = self._single_run(question_text, i + 1)
 
             yield result
 
             # If we've reached the confidence threshold, buzz and stop
             if early_stop and result["buzz"]:
+                if i + 1 < len(question_runs):
+                    yield self._single_run(question_runs[-1], len(question_runs))
                 return
 
 

src/workflows/structs.py

@@ -307,7 +307,7 @@ class Buzzer(BaseModel):
     """Configuration for when to buzz in a tossup question."""
 
     method: BuzzerMethod = BuzzerMethod.AND  # Logic to combine thresholds
-    confidence_threshold: float = Field(default=0.8, ge=0.0, le=1.0)  # Minimum confidence to trigger a buzz
+    confidence_threshold: float = Field(default=0.5, ge=0.0, le=1.0)  # Minimum confidence to trigger a buzz
     prob_threshold: float | None = None  # Optional log probability threshold
 
     class Config:
@@ -345,7 +345,7 @@ class Buzzer(BaseModel):
 class TossupWorkflow(Workflow):
     """Workflow specialized for tossup questions with buzzing capability."""
 
-    buzzer: Buzzer
+    buzzer: Buzzer = Field(default_factory=Buzzer)
 
     def get_answer_model(self, answer_var: str | None = None) -> str | None:
         answer_var = answer_var or self.outputs["answer"]