You seem to be viewing a Friendly-to-them singularity as freezing in place the couple's utility functions. I agree that it might be able to stabilize it against currently-known changes, such as those you cite, fading limerance, human pair bond stability, some others. I'm skeptical about stability with respect to all important changes over a million years. Even a superintelligence is going to encounter surprises, whether from exploration of the boundaries of design spaces or exploration of physical space. Even for it, the future is uncertain - and the balancing of subgoals and values must likewise have some uncertainty. If the consequences of one of those surprises makes one or both of the members of a couple morph into something rather different, is sticking with the original bond sensible, or even meaningful? If the couple precludes all such changes, is that at all a reasonable choice over such a long time period? Is it even viable? Precluding change in the face of surprise is a dangerous choice.