Once a weather API is deployed, what do you actually keep an eye on? Uptime alone feels too limited, especially if responses are technically successful but the data quality starts drifting. Which metrics have helped you detect problems before users noticed them, and which ones turned out to be less useful than expected?