You need to bear in mind that only one transformation rule is ever executed for each dimension source value in an imported data record. This means that you cannot have a star-to-star mapping to map your valid members as this will stop your wildcard catchall mapping from ever executing, as alluded to by the previous replies you have had. In this use case you would probably have to have a whole load of one-to-one maps just to pass through the valid values. This is probably not the best approach as it could be time consuming to setup (depending on how may members you have in your UD2 dimension) and would involve constant ongoing maintenance. Mapping by exception, if possible, will result in less map maintenance in almost all circumstances.
A much more streamlined solution would be to make use of OneStream Event Handlers, as suggested by rhankey, to either Bypass those records and send an alert telling the user / admin to address the missing metadata issue or in your case (where the source and target values should be the same) the optimal solution would be to create those new target members on the fly in your OneStream data model so your source data can immediately be validated and loaded to your target model. This would eliminate the need for any manual intervention / maintenance as part of that integration process.