I think I know what may have happened.... if anyone can shed light I'd appreciate it!

So the correlation matrix directly via CONN exports as ROI-by-ROI, as even when I select the relevant ROIs (ie 36-49 for CO), it groups them together in the designated order but still ROI-by-ROI.

The reason I couldn't group these ROIs is because the participants were preprocessed in two different projects (see photo for example of first round of participants' matrix). 

However, to do it outside of CONN & we wanted just network-by-network, which would by 6x6 (looking at 6 networks, without all the ROIs), couldn't do that from CONN so I tried manually. Therefore, when I averaged the values, it wasn't accurate at the ROI-by-ROI level, therefore the diagonals were not=1 (ie correlated with itself).

Thanks for your input!