Anyhow, thanks for the pointers and I'll take a look at Sofia and report
something to them if need be. I think my first option may be to disable
session timers on the test route initially, just to see if there's anything
else afoot.

Question: are they even willing to look at their SIP handling? Seriously, they are bombing out a call because they don't want to ignore a header that they feel is meaningless but that others feel isn't meaningless? If everyone writing a SIP stack did something capricious like drop a call because the packet had a valid header that they just didn't want to see at that point in the dialog then SIP interop would be royal mess of voodoo programming and vendor finger pointing. It's a good thing that silly SIP implementations are rare. Oh wait...

-MC

P.S. - As to digging into Sofia - you are either brave or foolhardy! ;) You might want to talk to the Sofia devs directly and ask them about this exact scenario. They might be able to save you some time, or at least guide you to the right spot in the code.