Instability in detection of aruco board

I am performing the detection of a board composed of 3 arucos in the shape of a triangle (each marker is placed in the center of each side of the equilateral triangle). I have decided to use this resource in order to be able to estimate the position of the board even if there are occlusions.

The problem comes when one of the markers is covered, since the reference system changes slightly, providing incorrect X and Y rotation values. That is, let’s suppose that when the three markers that make up the board are displayed, the rotations are zero, but when one of the markers is covered, the Z axis moves a little, generating a rotation value of between 2 and 5 degrees, even though there is no such rotation.

I am thinking that the error may come from lack of accuracy in defining the points of the markers that make up the board in three-dimensional space or maybe I should adjust some parameters of the detection to improve my results.

Thanks in advance for your answers, if necessary, I will upload the code and videos/images to better contextualize the problem I am facing.