Four log levels, with a rule for each
error means a person should look at it today. warn means something degraded and recovered on its own, and deserves a count on a dashboard. info tells the story of a request, one line per meaningful step. debug stays off in production.
If error fires a hundred times an hour on a healthy system, it no longer means anything. Downgrade things until the error log is quiet on a normal day, then alert on it.
observability