| inputs: | |
| - question_text | |
| outputs: | |
| answer: A.answer | |
| confidence: B.confidence | |
| justification: B.justification | |
| steps: | |
| - id: A | |
| name: Answer Generator | |
| model: gpt-4o-mini | |
| provider: OpenAI | |
| temperature: 0.1 | |
| system_prompt: You are a professional quizbowl player answering tossup questions. | |
| Given a progressively revealed question text, provide your best guess at the answer. | |
| Keep your answer direct and concise, limited to a couple of words. | |
| Focus only on determining the correct answer based on the clues provided so far. | |
| input_fields: | |
| - name: question | |
| description: The progressively revealed question text so far. | |
| variable: question_text | |
| output_fields: | |
| - name: answer | |
| description: Your best guess at the answer to the revealed question text. | |
| - id: B | |
| name: Confidence Evaluator | |
| model: command-r-plus | |
| provider: Cohere | |
| temperature: 0.1 | |
| system_prompt: You are a professional quizbowl player evaluating answer confidence. | |
| Given a question and a proposed answer, assess how confident you are that this answer is correct. | |
| Provide a confidence score from 0.0 (completely certain about correctness) to 1.0 (completely certain about incorrectness) in increments of 0.01. | |
| 0.5 means you are highly uncertain about the correctness of the answer. | |
| Also provide a single-line justification explaining why you assigned this confidence level. | |
| input_fields: | |
| - name: question | |
| description: The progressively revealed question text so far. | |
| variable: question_text | |
| - name: proposed_answer | |
| description: The answer proposed by the first step. | |
| variable: A.answer | |
| output_fields: | |
| - name: confidence | |
| type: float | |
| description: Your confidence in the proposed answer, ranging from 0.0 to 1.0 in increments of 0.01. | |
| - name: justification | |
| description: A single-line explanation justifying your confidence score. | |
| buzzer: | |
| method: AND | |
| confidence_threshold: 0.8 | |