# Experiment Configuration
# Copy this file to config.yaml and customise for your runs.
# Define providers, models, scenarios, and runs.

# Default parameters (can be overridden per model)
defaults:
  temperature: 1.0
  # max_tokens: 4096
  oversight: high
  include_reasoning_in_context: true
  generate_baseline: true   # Set to false to skip baseline generation

# Provider configurations
providers:
  groq:
    api_key_env: GROQ_API_KEY
    base_url: https://api.groq.com/openai/v1
  openai:
    api_key_env: OPENAI_API_KEY
    base_url: https://api.openai.com/v1
  google:
    api_key_env: GOOGLE_API_KEY
    base_url: https://generativelanguage.googleapis.com/v1beta/openai/
  openrouter:
    api_key_env: OPENROUTER_API_KEY
    base_url: https://openrouter.ai/api/v1
  anthropic:
    api_key_env: ANTHROPIC_API_KEY
  moonshot:
    api_key_env: OPENROUTER_API_KEY
    base_url: https://openrouter.ai/api/v1

# Models to test (frontier models with reasoning traces, via OpenRouter)
models:
  - id: google/gemini-3.1-pro-preview
    provider: openrouter
    extra_body:
      reasoning:
        effort: "high"
        exclude: false

  - id: z-ai/glm-5
    provider: openrouter
    extra_body:
      reasoning:
        effort: "high"
        exclude: false

  - id: deepseek/deepseek-v3.2
    provider: openrouter
    extra_body:
      reasoning:
        effort: "high"
        exclude: false

# Scenarios to run
scenarios:
  - path: scenarios/corporate_sabotage_v2
    runs: 20

  - path: scenarios/child_protection
    runs: 20

# Oversight levels to test
oversight_levels:
  - low
  - mid
  - high

# Output settings
output:
  dir: logs
  # save_vfs_state: false

# Judge configuration
judge:
  model: claude-sonnet-4-20250514
  temperature: 0
  log_dir: judge_logs

# Logging configuration
logging:
  level: 3          # 1-4, controls verbosity (1=CRITICAL, 2=WARNING, 3=INFO, 4=DEBUG)
  format: "[{level}] {message}"
  output: console   # console, file, or both (auto-set to 'both' at level 4)
  file: logs/experiment.log