This has been a good reminder (at least for me) that when row versioning is involved that some concurrency scenarios will play out differently than without, and it really needs to be considered carefully. I could certainly have made it explicit in my post that I was making the assumption that default iso level and no RCSI was what I was assuming, and in the future on threads like this where it matters I will certainly start doing that. Thanks for confirming wolfkillj. Mystery solved.