This is going back to the depths of time for me, so apologies if I haven't remembered this correctly.
There are essentially two processes available to aid in developing rules - inductive, and deductive.
The inductive processes are the ones where you produce rules based on observations of recurring patterns. IMHO, all rules respecting composition are inductive in nature.
A rule formulated using induction is, paradoxically, both more useful, and less likely to be "true" than one arrived at by deduction. More useful, because it is likely to incorporate more observations, but less likely to be "true" because it is based more on likelihood, than observable certainties.
Matrix metering is essentially an inductive process.
There is a very large likelihood that if the composition of a photograph is satisfying, than it is consistent with other photographs that previously were deemed to be satisfying.
You don't have to be aware of the rules themselves, to be aware of the images that are consistent with them.
You probably do have to be aware of them, however, in order to effectively break them.