Well, the paper makes no mention of any separate hidden traces. These seem to be just direct answers without any hidden thinking tokens. But as the thinking part is just a regular part of the generated answer I'm not sure it makes much difference either way.