An Auto Scaling group processes SQS jobs. Average processing time is 10 seconds, and acceptable average queue wait is 120 seconds. CPU utilization is low because workers wait on external I/O. Which TWO choices provide an appropriate scaling signal and initial target? (Choose TWO.)

1. Scale on average CPU utilization alone because low CPU proves no backlog exists.
2. Use visible queue messages divided by the number of in-service worker instances as a backlog-per-instance metric.
3. Use an initial target of approximately 12 visible messages per worker, then validate it against observed workload behavior.
4. Scale down to zero whenever the queue has messages but average CPU is below 20%.
5. Set the backlog target to 1,200 messages per instance by multiplying wait time by processing time.