Thanks Rick! Didn't even think about degrees of freedom. Makes so much sense.

The short story here then is that we cannot do motion correction if we for each block want to use the same number of TRs.
I could extend the blocks to each contain data for about 14 extra TRs (include the TR's left of the first red block until 0). Making the snipped 27 TRs long. I shoulnd't bother with testing that then if you say 50.

Is there another way to correct for motion in this scenario?

We can always run the regression on thefull dataset after the run is complete to check if the difference between just looking at the affective image (first image) and trying to down regulate (second time they see the image) is still there. To calm reviewers that they didn't learn motion, instead of actual down regulation.

=)