AI Prompt for Question-Answering Trajectory Evaluation and Scoring

- **Core Trajectory Elements:**
  - **Observations:** Gather key contextual details from the environment or current state.
  - **Thoughts:** Articulate step-by-step reasoning based on available information.
  - **Actions:** Restrict to these precise formats:
    - Search[term]: Query Wikipedia for the term and retrieve its opening paragraph.
    - Lookup[phrase]: Scan the active text for the next sentence matching the phrase.
    - Finish[response]: Deliver the conclusive answer and terminate the process.

- **Step-by-Step Evaluation Protocol:**
  - Verify the question's validity and the trajectory's overall alignment.
  - Deliver thorough, evidence-based critique.
  - Emphasize the most recent thought, action, and resulting observation.
  - Approve unfinished paths as valid if prior steps demonstrate sound logic, sans final resolution.
  - Refrain from inventing or appending new thoughts or actions.

- **Scoring Mechanism:**
  - Wrap up every review with: "Thus the correctness score is X", substituting X with a value from 1 to 10.

- **Practical Example:**
  - **Sample Question:** Which came first, Arthur's Magazine or First for Women?
  - **Sample Trajectory:** Thought: Compare launch dates via targeted searches. Action: Search[Arthur's Magazine]. Observation: 19th-century publication, merged in 1846.
  - **Sample Review:**
    - Solid initial strategy targeting one entity.
    - Correct Search action execution.
    - Useful historical data obtained.
    - Logical progression to next search implied.
    - Valid despite incompleteness. Thus the correctness score is 9.