Am 06.03.21 um 23:14 schrieb Allin Cottrell: Ah, I think understand the problem now. So the "default" data type of a 1-by-1 matrix is "scalar" as long as the data type is not explicitly declared, e.g. as in <matrix x = b.m>. I wanted to do some assertion of individual vector values. Here is a simplified toy example (in my application I have a more general architecture avoiding those two loops) but it illustrates the "issue". So, the point is in looping through all vector values by indexing each value. However, indexing a scalar (of what I though is a matrix) is (of course) not possible. Thus, the script fails here. The solution is to test the data type of "m*" and deciding whether one needs a loop or not. <hansl> clear set verbose matrix m1 = mnormal(1, 1) matrix m2 = mnormal(2, 1) bundle B = _(m1, m2) matrix expected_m1 = {-3.2} matrix expected_m2 = {-3.2; 5} print "Compare m2 values" loop i=1..rows(expected_m1) assert(B.m2[i] == expected_m2[i]) endloop # FAILS as indexing for the actual scalar m1 is not possible print "Compare m1 values" loop i=1..rows(expected_m1) assert(B.m1[i] == expected_m1[i]) endloop print "finished" </hansl> But I don't want to push this too far. It the spec is currently such that <x = b.m> becomes a by default a scalar (as long as not explicitly declared as a "matrix" type), this is fine as long as it is intended by the specs ;-) Artur