Since we weren't counting cached responses as successes, in Prometheus there were showing up as `unknown`, which contributes to the overall error budget.

We need to follow up with a runbook change that also counts cached responses as successes.