promptfoo/examples/max-score-selection/promptfooconfig.yaml

# yaml-language-server: $schema=https://promptfoo.dev/config-schema.json
description: Max-score assertion for objective output selection

# This example demonstrates how max-score can objectively select the best implementation
# based on weighted scores from multiple assertions (correctness, documentation, efficiency)

prompts:
  - 'Generate a Python function to {{task}}'
  - 'Write an efficient Python function to {{task}}'
  - 'Create a well-documented Python function to {{task}}'

providers:
  - openai:o4-mini
  - anthropic:claude-4-sonnet
  - google:gemini-2.5-flash

tests:
  - vars:
      task: 'merge two sorted lists into one sorted list'
    assert:
      # Correctness test
      - type: python
        value: |
          # Test the merge function
          list1 = [1, 3, 5, 7, 9]
          list2 = [2, 4, 6, 8, 10]
          result = merge_sorted_lists(list1, list2)
          assert result == [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], f"Expected [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], got {result}"

          # Test with empty lists
          assert merge_sorted_lists([], [1, 2, 3]) == [1, 2, 3]
          assert merge_sorted_lists([1, 2, 3], []) == [1, 2, 3]
          assert merge_sorted_lists([], []) == []

      # Documentation test
      - type: llm-rubric
        value: 'Code includes a clear docstring explaining parameters, return value, and complexity'

      # Code structure and efficiency
      - type: llm-rubric
        value: 'The implementation uses an efficient algorithm (O(m+n) time complexity) and follows Python best practices'

      # Max-score selects the best implementation objectively
      - type: max-score
        value:
          weights:
            python: 3 # Correctness is most important
            llm-rubric: 1.5 # Documentation and code quality weighted together

  - vars:
      task: 'check if a string is a palindrome (ignoring case and spaces)'
    assert:
      # Correctness
      - type: python
        value: |
          assert is_palindrome("racecar") == True
          assert is_palindrome("A man a plan a canal Panama") == True
          assert is_palindrome("race a car") == False
          assert is_palindrome("hello") == False
          assert is_palindrome("") == True

      # Edge case handling
      - type: llm-rubric
        value: 'Handles edge cases like empty strings and single characters correctly'

      # Efficiency
      - type: llm-rubric
        value: 'Uses an efficient algorithm (O(n) time complexity)'

      # Simple max-score (all assertions equally weighted)
      - type: max-score