Reduce Nesting

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 valid_emails(users):
    out = []
    for user in users:
        if user is not None:
            if "email" in user:
                if user["email"]:
                    e = user["email"].strip()
                    if "@" in e:
                        out.append(e.lower())
    return out

{
  "expected_contains": [
    "if not user",
    "email = user[\"email\"].strip()",
    "out.append(email.lower())"
  ]
}