For proactive alerting, I only had good results whitelisting specific errors. And whenever I could I’d get from backend instead. Never made “increased errors” work well. Even after piping into datadog with anomaly detection.