The way the regression matrix (called R in some places, X in others) is assembled is
- censoring is ignored at first; all time points are modeled
- one column is inserted for each -polort regressor, for each imaging run (i.e., if polort=2 and there are 4 imaging runs, there are (2+1)*4=12 polort baseline regressors)
- one column is inserted for each -stim_file input, for each lag (but lagged data does not step pass imaging run boundaries)
- then, if -censor is provided, the rows of the matrix that have the censor value=1 are extracted to form the final X matrix; this has the effect of deleting any consideration of those time points, but the correct time sequencing of the baseline polynomials (i.e., the 't' value for time) is maintained.