As far as this goes, I agree that it makes sense to be flexible and consistent with earlier specification versions; i.e. to permit annotation of type parameters both on fields and their associated accessor methods.
With regard to the mechanism used for reaching the validated contained values, my feeling is that some combination of the proposals at [1] is correct (i.e. the explicitness of the original proposal plus the aspect of your proposal to combine the extracted value with the associated path/node). It would probably also be helpful to provide an annotation-based shorthand for class developers to mark the extracted elements of custom container types without having to write an extractor implementation (which ties into my otherwise clueless response on an earlier thread ;) ).
Matt