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 positive(nums):\n    out = []\n    for n in nums:\n        if n > 0:\n            out.append(n)\n    return out\n

{
  "expected_contains": [
    "[n for n in nums if n > 0]",
    "if n > 0"
  ]
}