The exact version is 16.5s013. I don't know if there is any bug correction between our version which could explain the difference.
Actually, the extraction process works fine for some other differential pairs on the board. And considering the ones which generates errors, SigXplorer returns the complete transmission line anyway, but the impedance is not calculated for all the segments of the nets.
It looks like it is more a design setup issue, am I wrong ? Actually the company I work for (as an intern) has a PCB routing team, but they don't perform any simulation. These are made by the engineering team afterwords. So maybe the routing team did miss some parameters for some nets which could explain the problem...?
Might it be linked to a bad model assignment, stackup setup or anything like this ?
As the "set" command returns many things linked to the company servers and licenses locations, I'm not sure I'm allowed to post it... But as the extraction process works for sone nets, I think that all the paths are well defined (automatic installer, specially designed for the company).