So the design of Param Role goes through all the trouble of making the params full-blown Moose Attributes rather than just arguments, and a parameter object rather than a hashref, but then you are careful not to close over either of them but only on plain values extracted from them.
The docs show it would be my $something= $p->something. I didn't look to closely before, but now I'm not sure... I guess the parameter object $p is a Moose object that "has" the parameters specified, so ->something is the normal attribute getter for a ro attribute.
Ref-counting leaks would be some kind of circular reference that the system is not tracking correctly. Or for whatever reason, the past existence of such leaks is why you are leery of closing over the complicated linked-up Moose objects involving anonymous classes.
If I don't have the same penchant for making everything look like an accessor sub, declaring a lexical variable to hold the value returned from the param's accessor would do the same thing. OTOH, with your accessor method, only that one method needs to be a "closure", if Perl cares whether a closure is actually referring to anything that it might be closed over.