[logback-user] Sanity Checking the Log Stream

I found an interesting problem while configuring our Amazon Kinesis Connector. After starting up our service everything looks fine, but after 15 minutes or so, all hell breaks loose. The Connector is Amazon's code, and it does not throw any exceptions that our code can catch, rather it just reports a lot of errors in the log stream. This is clearly an insane situation that we need to catch, so I was thinking of writing a custom appender, filter, or turbo-filter to monitor the log stream for insanity. For example, errors per minute over some threshold. Does this seem like the right way to address this problem? Have other people used similar techniques, or other techniques to sanity check the log stream? Cheers, Eric