Simulink "Sum" block giving wrong result

I ran a simulation that produced an unnexpected result. It was supposed to be just a noise floor, but for some periods this signal seems to stay constane and then spike before returning to normal. And it happened twice in this particular length of simulation:

I tried the debugger and found an error in a sum block. I've tried re-running multiple times, the error occurs at the same timestep (my noise input didn't change between these runs). If I change the tolerance, it still occurs, but at a different time step.

Example, in this step, the result of the sum of this two signals is correct 1.9794e-11 - 1.905161e-11 = 7.43e-13;

As well as in the next step, -9.639417e-13 + 3.0406e-12 = 2.07e-12

however, at the next step, the signals are updated, but the sum, wrongly, still displays the previous value:

It continues for several steps. If I choose to run the remaining of the simulation, eventually it just "explodes" after a while and goes back to normal.

What could be going on?

[Youssef Noureddine]

Hi Arthur,

Can you log the value of the three signals that go into the add block and save it to a .MAT file and attach it here so that I can take a look at it and try it on my end? I can take a look at it tomorrow, but having the signal value history will greatly assist me in determining what could be causing what you are encountering.

Sincerely,

Youssef

[Arthur Reis]

Sure thing.

If I import them into matlab and try to sum them, the result is different from what simulink has calculated.

input1 = load('input1.mat').data;
input2 = load('input2.mat').data;
input3 = load('input3.mat').data;
output_from_add = load('output_from_add.mat').data;
input3.Data %input3 is a constant 0;

ans = 0

sumofinputs = input1 + input2;
df = sumofinputs - output_from_add;
plot(df);

[Arthur Reis]

~~Just an update: selecting ode113 as a solver seems to solve this problem (ode45 is the problematic). And I haven't tried the others.~~

~~So, this could be related to the solver selection. Any insights of why?~~

Edit: Ignore the previous lines... it just changed the timestep where the error occurs