experiments.yaml

# Experiment suite configuration
# Run with: python run_suite.py experiments.yaml

suite: multi-repo-v1
description: >
  Cross-repository evaluation of Graph-Manager vs RAG baselines.
  Each run evaluates all 6 methods (GM/RAG × progressive/baseline + 2 raw RAG)
  on the same issue set per repo.

defaults:
  n_issues: 10
  max_turns: 6
  # Optional per-agent defaults (fallback to max_turns if omitted)
  # manager_max_turns: 6
  # rag_max_turns: 6

experiments:
  # --- Flask (small web framework, 11 issues in SWE-bench) ---
  - repo: pallets/flask
    source_prefixes: [src/flask]

  # --- Requests (HTTP library, 44 issues in SWE-bench) ---
  - repo: psf/requests
    source_prefixes: [requests]

  # --- Pytest (testing framework, 119 issues in SWE-bench) ---
  - repo: pytest-dev/pytest
    source_prefixes: [src/_pytest, src/pytest]