After having this task languish in my todo for years, at least, I have finally reduced my goal to the important 90% and applied some JFDI.
Pod::Coverage::TrustPod acts like Pod::Coverage::CountParents, but accepts non-whitespace lines inside "Pod::Coverage" POD targets as "trustme" instructions.
In otherwords, given this module:
package Foo;
=begin Pod::Coverage
bar
=end Pod::Coverage
=head2 baz
This routine does stuff.
=cut
sub foo { ... }
sub bar { ... }
sub baz { ... }
sub _xyzzy { ... }
Pod::Coverage::TrustPod will return 2/3 coverage. _xyzzy is private, so uncounted. There are docs for baz, bar is trusted (because it's in the Pod::Coverage =begin block) and there are no docs for foo.
The basic use case is that now my xt/pod-coverage can now read:
all_pod_coverage_ok({ coverage_class => 'Pod::Coverage::TrustPod' });
and the information on what code is written but undocumented can be put where it belongs -- near the code that is written but undocumented. This will help eliminate both (A) marking a symbol trustme/also_private in all of a dist's packages so that I can use all_pod_coverage_ok and (B) having to write a bunch of pod_coverage_ok lines to avoid problem A.