One thing that surprised us: most cost regressions aren’t “big changes”.

They’re small PRs that: - Add always-on resources - Increase replica counts - Introduce N+1 API calls or polling loops

By the time dashboards show it, the code is already in prod.

Happy to answer questions on how the estimates are calculated or what patterns we flag.