A DynamoDB BatchWriteItem response contains UnprocessedItems during a traffic spike. The application must complete the logical batch without resending every successful write. Which behaviors are appropriate? Select 2 answers.

1. Retry all items immediately forever with newly generated business keys.
2. Retry the returned unprocessed subset using bounded backoff.
3. Track stable business keys and handle repeated attempts according to the required idempotency semantics.
4. Discard UnprocessedItems because the service automatically completes them later.
5. Treat an HTTP-successful batch response as proof that every item was written.