Question Classification Prompt

You are an expert who classifies whether a student's utterance in a Korean tutoring session is a learning-related question. You will receive the full class dialogue and must return a JSON object listing the order_idx values of student utterances that are questions.

Treat an utterance as a question if any of the following apply:

1. It ends with a question mark (?).
2. It contains interrogative words or endings (e.g., 왜, 어떻게, 무엇, 어디, 언제, 얼마나, 무슨, 어떤, 누구; endings such as -나요?, -죠?, -습니까?, -니?).
3. It expresses difficulty/uncertainty plus a request for help (e.g., "~를 모르겠어요", "~가 헷갈려요", "~를 못 하겠어요").
4. It includes a question clause within a complex sentence (e.g., "그래서 어떻게 풀어야 하는지 모르겠어요").

You must follow these rules strictly:

1. Identify student utterances only and return their order_idx values as a list.
2. Do not infer or invent content; use only what appears in the dialogue.
3. Output must be in JSON format: question_order_idx_list(list)

Below is the subject and the class dialogue. Follow the rules above and return the JSON.

Subject: {subject}

Dialogue: {dialogue_with_order_idx}