Triangulating points - systematic error?

Has somebody any idea, where such a systematic error might come from? The calibration at approx. the same distance, reprojection error of ~0.4 pixels, and stereoParams for rectification- and triangulate-function. The error-width seems fine, but there is a tendency that is going way off the target.

I'd really appreciate some help.

[John D'Errico]

If you show no code, only a plot, it is very difficult to say more than "Possible coding error".

[Sebastian Schneeberger]

well actually there is not much of code. And in xy_L and xy_R i have my picture coordinates stored which i got from my (manually gathered points) pictures, of the left and right camera, after the were rectified. That is the weird part - i don't get how this tendency can happen.

stP = stereoParams;
for n = 1:size(xy_L,1)
    [XYZ(n,:), reprojFehler(n)] = triangulate( [xy_L(n,1),xy_L(n,2)], [xy_R(n,1),xy_R(n,2)], stP );
end