With 'TENT(0,15,7)' as your HDR model for -stim_times_AM2, you have totally 14 beta values for each condition. The first seven of the 14 betas are the effect estimates corresponding to the average RT across trials of the condition for the subject, and the second set of seven effect estimates is the linear modulation effects of RT.

If you only care about the BOLD response of a condition with lags of 2-4 corresponding to the average RT, you may consider either

-gltsym 'SYM: CondA[[2..4]] -CondB[[2..4]]' -glt_label 1 CondA-B

or the area under the area (AUC) approach

-gltsym 'SYM: CondA[2..4] -CondB[2..4]' -glt_label 2 UAC_A-B

See more details here:

[afni.nimh.nih.gov]

Gang Chen
[afni.nimh.nih.gov]