I think you are spot on regarding the subcortical structures.
The column forming neurons have reciprocal projections to subcortical structures that don’t seem to show up anywhere in the HTM theory. Some of these connections are prominent and the key to the top-down learning path and the related error distribution in this model of the cortex.
I see the RAC as a mechanism for gating the sensory stream regulated by how well the higher structures “resonate” with what is being sensed. If we have seen this and recognize it (ho-hum) not much gets through. If there is confusion with a lot of bursting going on (what the heck is this?) then the floodgates open up, driving enhanced learning. Call it attention if you want. Crick did.
http://www.pnas.org/content/81/14/4586
From what I have read, RAC fits into the spectrum of several mechanisms that modulate learning rates from single shot to pure slow Hebbian.
For that matter - there is a zoo of cell morphologies in the cortex that don’t show up anywhere in the HTM cannon. I am working to reconcile some of these with the HTM model.