Hi Michael,
Most regressors should have discontinuities at the
run breaks.
The stimulus regressors should break between runs
because BOLD responses do not cross runs. The
polort regressors should model the drift per run
for a similar reason, slow drifts or fast, they
are still discontinuous across run breaks. The
motion terms can be modeled either way (depending
on the options given to afni_proc.py, for example),
because at least they coincide with the TRs.
Let 3dDeconvolve know about the run breaks.
- rick