The problems of any input/output-gate macro, stand to contaminate their containers, too. If a subnetA contains a subnetB, and if that subnetB starts with any of the entrance-macros "or," "xor," "priority," then subnetA itself is very probably problematic, because it is well-nigh impossible to manage the quirks of those input/output macros. e.g:
The output macros do not match the input macros (other than the non-exception of the ordinary Petri net join/fork, i.e: Whatever copycat82 itself implements, leads to mismatches. It is difficult, or impossible to imagine any reduction-heurisitics with such quasi-static (or, un-proper, to use Petri net terms) "logic" macros.
If a subnet is un-proper, it is not reduceable. And no ways to represent the deadness or non-properness, even if it may exist within the subnet - let alone any mechanism to specify in what sequences of tokens, such problems may occur. i.e: If a subnet contains problems, it is not known by the container net. To map the internal problems, to the already problematic input/output macros, does not appear possible (unless when the internal problems may be, trivially, at near an edge of the subnet, in which case, it may be carried out).
copycat82 is vague about what it means when it "cannot continue the analysis, if any of the net places is to contain multiple tokens." Does it mean to dump any model away, the moment at any point, the verifier encounters not 1-bounded'ness? Or, does it mean to imitate E-nets behavior, i.e: let-transition-not-fire-until-its-output-is-empty? Although, in words, the first appears the case, it must have been the latter, because otherwise, the entrance-macros themselves would suffice to dump most, if not all models. The "or" gets stuck, if tokens are not deposited in strict-alternation. The "++" with, or without priority, would overfill the place after it - by definition. Any subnet that starts with a "++," is necessarily doomed.
Especially the "or" and "priority" certainly do not do what their name would suggest, the "xor" is another source of stickiness once after it leads to a deadlock.
If any macro/subnet contains any of these, then that macro would very probably not be reducable. If there is even a single loop, such an i/o macro would let the model to be dumped away, because the (non-deterministic) reachability test would find them, as it verifies exhaustively (and copycat82 does not continue, when such is found). It is without any of the versatility of Macro E-nets (NN73), although such volatile input macros of copycat82 lose the SARA preference of staticness (with the UCLA graphs, as the SARA GMB control graph), either.