Extract And Guard

The model receives the prompt (and optional system message). The run uses scorer contains_all 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.

def normalize_users(users):
    out = []
    for i in range(len(users)):
        u = users[i]
        if "name" in u and u["name"] is not None:
            name = u["name"].strip().lower()
            if name != "":
                out.append({"name": name, "active": bool(u.get("active", False))})
    return out

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