A custom metric uses a unique request ID and customer ID on every publication. Time-series count and cost rise sharply. Which changes are appropriate? Select 2 answers.

1. Remove all diagnostic correlation from every system permanently.
2. Assume CloudWatch ignores dimension values when identifying a metric.
3. Keep detailed correlation identifiers in appropriately controlled logs or traces.
4. Add a random deployment token to every datapoint to reduce cardinality.
5. Use bounded operational dimensions for aggregate metrics.