| # AgentDisruptBench — Benchmark Configuration | |
| # ============================================= | |
| # Top-level configuration for a benchmark run. | |
| # Use with: python -m evaluation.run_benchmark --config config/benchmark.yaml | |
| # | |
| # CLI arguments override any values set here. | |
| runner: rac # simple | openai | langchain | rac | |
| profiles: [clean, hostile_environment] # Disruption profiles to evaluate | |
| domains: null # null = all domains, or: [retail, travel, ...] | |
| max_difficulty: 5 # 1-5 | |
| seeds: [42] # Random seeds for reproducibility | |
| output_dir: results # Directory for report output | |
| verbose: false # Print agent reasoning to stdout | |