Cleanup

The model receives the prompt (and optional system message). The run uses scorer contains_any with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.

Refactor without changing behavior. Return only Python code.\n\ndef active_usernames(users):\n    out = []\n    for u in users:\n        if u.get('active') == True:\n            out.append(u['name'].strip().lower())\n    return out\n

{
  "expected_contains": [
    "for user in users",
    "user[\"name\"].strip().lower()"
  ]
}