A Lambda function processes Kinesis batches. Most records succeed, but one record repeatedly fails. The team wants supported partial-batch reporting and wants repeated delivery to avoid duplicate downstream effects. Which actions are appropriate? Select 2 answers.

1. Make downstream writes idempotent because retried stream processing can include records that were already handled.
2. Return arbitrary business labels as failure identifiers instead of the required sequence numbers.
3. Assume a successful write can never be repeated after partial-batch reporting is enabled.
4. Catch every exception and always report success without preserving failed records.
5. Enable ReportBatchItemFailures on the event source mapping and return failed records using the required sequence-number identifiers.