I don’t understand what this suggestion is trying to accomplish.
I think limiting guard, down to maybe 3 or 4, would be a positive change for Knight. This forces Knights to think a little more about who they guard rather than spam on King.
How does it accomplish that? Why is that desirable? You seem to want kings to have less reliable protection early on, but I’m not certain that that would really improve the game, and I definitely don’t think it would actually make knights deeper or more interesting to play (in practice, being able to only defend a few nights would leave knights with fewer decisions to make, not more, because on most nights they would do nothing.) If your goal is to make knights more fun to play, encouraging them to frequently do nothing is an… interesting way of doing it.
Mindlessly spamming defense on the king (outside of the first few nights, when it absolutely serves an important game design purpose by discouraging people from killing the king as soon as the game begins) is a terrible way to play a knight and isn’t encouraged by the current mechanics at all, since you don’t actually know if the king is on your side or not.
I think an additional change that could prove to make Knight feel more rewarding, is that sacrifice literally becomes a guarding ability, and Knight works like Merc, where he prevents visits, rather than killing himself. Knight should still be vulnerable to direct instances, but he prevents all visits to the target.
At this point it feels like you just dislike the entire concept of knights and want to replace them with a totally unrelated class.
Anyway, this feels like a bad suggestion that isn’t even trying to accomplish anything desirable. The basic protective abilities serve a vital purpose in the game by discouraging killers from going after “obvious” targets; it feels like this suggestion isn’t really considering the important roles that knights play in the meta.
If your goal is to make “predict when people will attack” part of the strategy required to play a knight, I just don’t see what it adds when they already have “predict who is going to be attacked” as a defining feature. In practice this would make it feel more random and would encourage killers to risk potshots at obvious targets more freely, which I don’t think would lead to interesting games.