Our analyzer currently works this way:

• recommended rules emit a diagnostic error
• non-recommended rules don't emit any diagnostic

This behaviour is different from the majority of linters/analyzers out there, and it's really confusing for people who use other tools. We intended to change it, so:

• lint rules will always have a severity (it will be Severity::Information by default)
• recommended rules won't change the severity of a rule

This means there could be lint rules that will have Severity::Warning rules and be recommended.