Perhaps we're double encoding when we shouldn't be?

Yes, my understanding after reading the specification again is that it should not be encoded twice.