Hi renevg,
I know how frustrate can be when no one answers on your message and you hope (or thought) that this is the right place to post questions or so.
Believe me or not, but before few days I found that someone post questions on this forum, and later he or she - replies to himself or herself cause it found solution to issue that has not been solved at the start. This is so SAD...but it's true... that people used to 'selftalk'.
Now, from my introduction, let's get to your issue.
I can say that, from my experience I first "set up" coordinate system in order to avoid any problems either with check points precision or with scale bars.
For setting up coordinate system, it's local coordinate system, I use 3 points which lays on flat surface (it can be floor, it can be surface of the table - anything that have flat surface in your model).
Then, I use created excel table, which I can share with you in order if you still face with that problem. From this 3 points I create custom coordinate system: first point is origin of that system, second point defines X axis and third point defines Y axis from that origin. You don't need 4.th point for defining Z axis cause Z axis is perpendicular to XY plane. If you measure distance between these points, excelent You can then use your measurements as values for each point which defines custom coordinate system. I did not mention, but these 3 points should form right-angle triangle, but If you can't predict which would fall 3.rd point - I also defined in excel table calculation of that 90° spot for 3.rd point. You can also done that for yourself if you know linear algebra.
NOW...Comes KEY PART once you determine coordinates of points defining your coordinate system you SHOULD check these points with check marks then hit UPDATE/TRANSFORM.
After this - You can freely add as much as possible scale bars, and if you have good measurements and good reconstructed 3D model there's no fear for bad results or don't know possible Agisoftmistakes.
Issues occures when you first define scale bars, then coordinate system, cause I'm not expert but I might think that issue lays - that Agisoft already give these points values without applying scale bars. When you apply scale bars then points don't colsolidate well, and that's the reason for high errors.