Kendeepp

Agree — a silent partial failure is worse than a hard fail. I alert on fill-rate drops too, not only on failed runs; a green run with half-empty rows is still a broken delivery. Thanks for the tip.

Thanks — that's exactly the tradeoff I keep seeing. A predictable bill usually beats a slightly cheaper unit price that buyers can't estimate before a run. One event per delivered result has been the simplest model for me to explain in a README.