It sounds like each run should have only 1 main task in
it (i.e. either vis or nv), yet the afni_proc.py commands
suggest both tasks are in both runs.

Would you show what the first few elements of each line
of the timing files are?


Theoretically, if there was one task in the first run
and a different task in the second, then analyzing either
way should produce very similar results. But there may
be small differences:

1. The EPI registration bases are a little different,
which will have a small effect on alignment. One would
hope the motion parameters for the run 2 would be akin
to just shifted versions between the 2 methods, having
similar derivatives, but they would not be identical.

That means that the motion regressors and the actual
voxel time series would be slightly different between
the 2 methods.

This small point could be mitigated by using the exact
same EPI registration base. But maybe your concern is
only theoretical.

The use of -align_opts_aea and applying the same base
for registration should correct this small difference.


2. By default regression of motion parameters is across
all runs at once. That will also have a small effect
on the betas, depending on the correlation of motion to
signal of interest.

This difference could be mitigated by using
-regress_motion_per_run.


But neither of these points should have a big effect
on the results. Which brings back the question of
the stimulus timing.

Have you compared the ideal time series plots? The
single run plots should match the 2-run plots.

- rick