Picture an Erlang web-application server.

A request that causes a 500 error response to the client, should also produce an Error log output on the backend. The request crashed. But, because this is an actor-based system, that crash didn't affect anything other than the request. It was Fatal to that interaction, but not to the system.

A request that causes the entire server process to spontaneously halt, should generate a Fatal log message. That is never supposed to be something that happens. The system is built, on a number of levels, to ensure that it doesn't happen. But it'd be really great, if it does happen, to have enough information to know why.

Ideally, a Fatal log message is accompanied by a crash dump.