[aaronmu]

The best way I can think of is to aggregate errors over time, categorize them and build a health check around those metrics.