Eigenvalues of a triangel matrix with higher dimensions (here 29) have sometimes imaginary part, where they should not have

[Anatom70]

I create an integer random (Range [-100,100]) triangel matrix and let the Eigenvalues calculate with the Evd method.
Due to theory the Eigenvalues are the diagonal elements and should be all real numbers!
I checked this and sometimes (in 1 of 15 cases) the Eigenvalues have an imaginary part:
Eigenwert 26 has imaginary part (41,0000000000697, 1,02305029205388E-05) on 29
13 0 0 0 0 0 0 0 0 0 0 0 0 .. 0 0
70 14 0 0 0 0 0 0 0 0 0 0 0 .. 0 0
-8 -39 75 0 0 0 0 0 0 0 0 0 0 .. 0 0
97 73 -18 81 0 0 0 0 0 0 0 0 0 .. 0 0
32 -90 -69 -96 -3 0 0 0 0 0 0 0 0 .. 0 0
-28 70 -99 -92 7 -64 0 0 0 0 0 0 0 .. 0 0
56 -5 47 26 -87 -96 -12 0 0 0 0 0 0 .. 0 0
9 -15 78 66 -79 3 71 -22 0 0 0 0 0 .. 0 0
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
-42 13 -21 -66 78 54 7 33 -33 -71 -8 47 76 .. 0 0
-76 81 14 -72 -24 -53 -10 -61 -47 -47 -46 -99 -10 .. 0 0
16 97 -76 37 -36 63 56 -92 -26 53 -26 -78 29 .. 41 0
-50 89 -22 70 1 24 3 -73 -9 52 -61 63 -86 .. 77 -75