Hi, somehow I want to believe that we discussed this before, but I can't find any reference to it and in any case I see the behavior in the Nov-14th snapshot. Consider this: <hansl> open denmark var 2 LRM LRY --quiet # starts at 1974:3 eval $system.t1        # 2 eval obsnum(1974:3) # 3 (correct) </hansl> And the same with $system.t2. But note that this only happens after 'var', not after 'system'! I've looked at the VAR_set_sample function in lib/src/var.c but saw no obvious bug there - but also I don't even know whether at the C level the v->t1 index is interpreted as 0-based or 1-based. Some loosely related observations on C code chunks that I looked at right now: -  We have this VAR_set_sample function and in lib/src/system.c there is a system_adjust_t1t2 function which basically seems to have the same aim (and internally calls list_adjust_sample from somewhere). Maybe some consolidation in the medium term might be reasonable. - In the equation_system_estimate function (also in lib/src/system.c) there is this part (lines 1604-6):    if (err) {        goto system_bailout;     }   and then after another block there is (lines 1621...):   system_bailout:     if (!err) {         sys->smpl_t1 = dset->t1;         sys->smpl_t2 = dset->t2;         if (!(sys->flags & SYSTEM_LIML1)) {             set_as_last_model(sys, GRETL_OBJ_SYS);         }     }     return err; Now I'm not too familiar with goto in C (wasn't goto declared taboo in the old days?) but to me this looks as if first err must be non-zero to jump  but then there's the opposite check for a zero err. So isn't the whole goto thing superfluous here and in the end just err is returned (if non-zero), or what am I missing? thanks sven