Hi Han,
I was thinking on a different design. The comparisons between the 4 conditions are done at the 1st level, for each individual. Same for the interaction between condidions as the experiment happens. At the second level, if I understood, the idea is to see if what was found (i.e., the COPEs 1, 3 and 5) from the first level are different than zero across subjects. No new comparisons between conditions or interactions at the 2nd level.
The simplest way to do this is to run 3 separate 1-sample t-tests, one for each of the COPEs that you are interested into. The reviewer wants F-tests, so you just define an F-test using the sole contrast of the 1-sample t-test.
Ideally, these 3 F-tests should be corrected for multiple testing, so in principle, you'd adjust the significance level to 0.05/3, although that could be a bit too stringent.
You may ask whether it could be all done all in a single big GLM. Yes, it can be done, with a design that'd be different than the one you suggested, with one exchangeability block per subject, sign-flippings for blocks as a whole, variance groups for each of the lower level COPEs, and correcting for multiple contrasts, but the implementation for that is not yet available.
All the best,
Anderson