A Kinesis consumer remains running without errors, but its output is increasingly stale. The team wants an alarm that reflects how far consumer processing has fallen behind the latest available records. Which signal is the most direct fit?

1. Only the producer IncomingRecords count, with no comparison to consumer progress.
2. Only the EC2 instance running state for the consumer host.
3. The applicable iterator-age or consumer lag metric, combined with processing throughput and error monitoring.
4. Only the number of successful consumer invocations, without their position or age.