[MENTION=42582]pemerton[/MENTION] you would have used RtD in [MENTION=463]S'mon[/MENTION]'s scenario or would you have said yes, having the wife been an extension of the player's control?
It depends on system and on "story" context. (Because I don't know all the details of S'mon's campaign, some of what I say may, to him, be obviously irrelevant to the particular situation that arose in his game.)
Re the second thing (ie context): a relationship that a PC is part of can either be colour/background, or can be a substantive aspect of gameplay. An example of the former: in my Prince Valiant game, two of the PC knights are married, but their spouses are just colour. The wooing and the weddings were big deals in play (one bigger than the other), but now that they've satisfied their dynastic ambitions they are planning to travel to Byzantium to fight Huns and join a crusade. A different sort of example: in my 4e game, two of the PCs have had familiars. The fact that a familiar is a semi-autonomous entity is, in 4e, just colur: the mechanical abilities conferred by a familiar are, in effect, bonuses conferred by a feat, and the fact that
in the fiction there is this other being involved is mere colour.
When a relationship is mere colour then I regard it as something for the player to look after unless s/he does something to bring it into the foreground (eg implanting the Eye of Vecna into one's imp familiar).
This doesn't mean that the relationship will always stay on an even keel. I have at least one player who can be very ruthless towards his own PCs if he thinks that's what the fiction demands.
When a relationship is not mere colour - of my currently active games, Burning Wheel is the main one where non-colour relationships are part of the game - then system becomes relevant.
In some systems, relationships are tagged/categorised as
inimical or
friendly (eg Burning Wheel) or as a flaw-type attribute or a buff-type attribute (eg HeroQuest revised; Marvel Heroic RP also supports something a bit like this). If a player chooses an inimical/flaw-type relationship then obviously all bets are off - and changing the nature of that relationship would require some significant success on the part of the player in the course of play.
If a player chooses a friendly/buff-type attribute, though, then the opposite applies: having that person turn into an enemy would only be the result of some significant failure in the course of play.
Because D&D (outside of the henchman mechanics) doesn't really treat non-colour relationships as an element of PC build, whether a relationship should be understood as friendly or inimical is going to depend much more on the fictional details; but that doesnt mean the classification is meaningless. Eg if the players succeed in a skill challenge to befriend so-and-so, then it would (in my view) be poor GMing to change the result of that outcome unless the players subsequently stake the relationship on some further outcome. Here's an example of that from my 4e game: the PCs befriended the baron; but then when they discovered his beloved niece was actually a Vecna-ite necromancer they informed him and asked him to deal with her as the law and justice required - the checks on that second occasion were not successful, and the outcome in the fiction was that the baron had an emotional breakdown and collapsed, thus signficinatly reducing the utility to the PCs of their relationship, and depriving the players of much of the benefit of their earlier success. (This example also shows that I regard it as perfectly good GMing, indeed obligatory!, for the GM to put pressure on the PCs' non-colour relationships - that's what drives a character-driven game. But
pressure isn't the same as unilateral turning.)
If I look at the Tanis-Kitiara example through the lens of 4e play, then I would imagine that Tanis's player establishes, at the start of play and as an element of PC backstory, that he has an ex-lover who had a dark-side streak. Trying to reestablish connections with her would then be an unfolding skill challenge; or perhaps, in other skill challenges, Tanis's player would declare actions that relate to or draw upon his connection with Kitiara.
A successful check in these circumstances might result in learning some bit of information about Kitiara's whereabouts, or receiving some aid from an unrevealed ally who is (presumably) Kitiara. The first failure in respect of these checks might be narrated as Kitiara having joined the baddies. The second failure might be that not only has she joined the baddies, but she's a leading dragon highlord. The final failure would reveal that she is, beyond a shadow of a doubt, neither a double agent, nor willing to betray the baddies to reunite with Tanis.
This more-or-less conforms to the procedure I described above: each of the player and the GM put up a possibility that falls within the bounds of plausibility and is the outcome that is salient for them; and then the check tells us whose possibility is the one that is actually realised in the shared fiction.