# Catch the problem nobody wrote an alert for

Maple learns how each service behaves at each hour of the day. When error rate, latency, traffic or one error type moves far outside that, it opens an anomaly with the numbers behind it.

## What Maple watches

### error_rate: Error rate

The share of failed requests per service and environment, against the same hours last week. A service needs 50 requests in the hour before it is judged.

### latency_p95: p95 latency

Slowdowns far outside the band and at least 50 ms slower. A nightly batch that is always slow is part of the baseline.

### throughput: Traffic loss

A busy service that drops to almost nothing. A slow Sunday does not count. Losing 95% of normal traffic does.

### error_spike: Error spikes

One error type firing far more often than usual in 30 minutes. Spikes on the same service are grouped and linked to their issues.

### log_volume: Error logs

A jump in error-level logs per service, for failures that never show up as a failed span.

### investigate: Investigated for you

Open an investigation from any anomaly, or turn on AI triage and an agent looks into each new one and attaches what it found.

## Related

- [Alerting](https://maple.dev/features/alerts.md)
- [Error Tracking](https://maple.dev/features/error-tracking.md)
- [Service Catalog & Map](https://maple.dev/features/service-catalog.md)
- [API Performance](https://maple.dev/use-cases/api-performance.md)