Current behaviour

As per https://support.atlassian.com/jira/kb/health-check-webhooks/ , 

The error health check uses a single threshold applied to the peak snapshot value within the window. There is no separate threshold for peak vs. average.

Suggested behaviour

The health check should be triggered by the average error ratio over the default time window (ie 1h), not just by the peak value.

Workaround

Currently there are no workaround for this behaviour.