# Minimal config to test LLM-as-a-judge evaluation feedback
# Usage: skydiscover-run initial_program.py evaluator.py -c configs/llm_judge.yaml

max_iterations: 10
checkpoint_interval: 5
log_level: "DEBUG"

llm:
  models:
    - name: "gpt-4o-mini"
      weight: 1.0
  temperature: 0.7
  max_tokens: 16000
  timeout: 120

  # Model used by the evaluator's LLM judge
  evaluator_models:
    - name: "gpt-4o-mini"
      weight: 1.0

search:
  type: "topk"
  num_context_programs: 2

prompt:
  system_message: "You are an expert programmer. Improve the given program."
  evaluator_system_message: >
    You are a strict code quality judge. Evaluate the given code and return a JSON object
    with scores between 0.0 and 1.0 for each metric. Be critical — only exceptional code
    should score above 0.8. Consider algorithmic correctness, edge case handling, and
    real-world production readiness in your assessment.

evaluator:
  timeout: 60
  max_retries: 1
  cascade_evaluation: false
  llm_as_judge: true

diff_based_generation: true
max_solution_length: 10000