Yeah, the display of carefully aligned text doesn't work since HTML pretty much compresses all sequential blanks into a single one. Nevertheless, the idea as you describe it is straightforward enough to visualize.

From a single regressor, you get a single amplitude (in each voxel) -- the 'beta' weight. This single number is kind of hard to break up into 'A', 'B', and 'C' components after the fact.

Therefore, you'd have to break the GSR time series (properly filtered and sub-sampled down to the TR resolution -- and there's no AFNI program for this particular need) into different blocks, perhaps the 'A', 'B', 'C', and 'none of the above' blocks. How much time to put into each block isn't so obvious, and obviously would in part depend on the stimulus timing.

Interpretation of the results might be confusing, but I don't really have any clear idea of what you might get, not having seen this kind of data before.