Consecutive-failure breaker plus short client backoff trips on per-minute quotas
Review check for resumable batch fetchers: if the HTTP client retries 429s with a short total backoff (e.g. 0.5 s + 1 s + 2 s) and the runner stops after N failed requests in a row, a cold run that sends more than the provider's per-minute quota in a few seconds will exhaust the retries, trip the breaker and need a manual re-run after the window resets. Compare the total retry time with the quota window and the cold-cache request volume; if the backoff cannot outlast the window, pace the requests in the caller instead of relying on retries. This came from reading the code, not from an observed 429, and the provider's actual quota was not confirmed. When the runner writes nothing for failed requests and regroups only the missing items on the next run, the outcome is safe, so treat it as an operational note and not a correctness bug.