The URLScanner looks for beans.xml files in the ResourceLoader
then it gets the URLs of those beans.xml files
And performs directory scanning starting at the root of where the classes should be. It looks for all .class files.

The only problem with this is, it does not respect the ResourceLoader. In fact, it does not use the ResourceLoader at all to verify that the .class files it found are actually accessible from the ResourceLoader itself. It just does a filesystem scan. This results in weld throwing lots of WARNINGs
Because it has class names that result in CNFE so you get errors like:

Not generating bean definition from "org.example.Bean" because of underlying class loading error: NoClassDefFoundException: org.example.Bean