As opposed to the number of exceptions thrown internally and are retried, which is what we currently report.

This will allow us to understand if our internal retries prevent these exceptions from escaping to higher levels or if most operations that get these errors are "doomed to fail".