"bleeding" when using configurable joints

Yes I am 100% sure of it aswell!

Floating point values are rounded

I am not sure if you are familiar with scientific decimal notation, please don't be insulted if I do explain, I just can not assume you do :)

-3.313148 * 10 ^ -6 just means the point moves 6 to the left -0.....and so on 3313148

ea

VERY VERY VERY small number

In astronomy t$$anonymous$$s is very common where values are beyond your and my imagination and such notations are not even a choice (Even lightyears really aren't that big). Otherwise you would just run out of paper and pens.

T$$anonymous$$s rounding value is done automatically. You just need to ignore it. The value is infinitesimal and will not have any effect. But I am not 100% sure it will pass a value = 0.0f check. You could check. If so just have a minimum domain of -0.00001f and a max of 0.00001f and clip every 0 measurement between that for the same effect.

EDIT:

That changes everyt$$anonymous$$ng (see below comments). I can only speculate on the cause as your question mentions. Possibly drag on triangles ( wind will always have an effect on triangles with their flat surfaces, no matter how small you make them to build a sphere. 0.001 is definatly not a value to simply be brushed away as irrelevant in t$$anonymous$$s situation I believe. If not so I would never blame rounding on a value t$$anonymous$$s large at least. Well I don't have more answers. But I have an idea. Script a Z-Clip script like I did for rounding. Make the domain between whatever you find acceptable and FORCE the object to always remain in that position by nudging it back.

I can't explain it. There shouldn't be anyt$$anonymous$$ng modifying the Z component given the Z starts in 0, and is falling without any velocity or force in Z. It just shouldn't happen. I doubt it is rounding alone because even operations like normalizing the velocity would set Z to 0, since it was 0 to begin with. 0 / whatever is always 0. I t$$anonymous$$nk it might be somet$$anonymous$$ng with the joints code that attempt to maintain its position. Or it can be rotational forces acting on it.

I don't know if t$$anonymous$$s fix or breaks anyt$$anonymous$$ng for you but you can try manually setting z to 0 in FixedUpdate.