That thought occurred to me as soon as I posted this. It’s not really an “unexpected behavior,” as you can achieve lots of weird effects from using really big numbers throughout the API. Lighting and part reflectance come to mind.
So on second thought, I don’t think this is a problem that needs to be addressed.
EDIT: I suppose if anything, the assembly should be destroyed. That’s what happens when you perform it on the server.
We will still investigate. We try to catch NANs before they can affect the simulation system and degrade it, but checking for NANs in some cases isn’t free. We can see if this is reasonable to sanitize.