After some exploring:

Just the presence of a functional WOT score alone is enough to evaluate your own valence of someone because you end up building a neighborhood of users with scores you're comfortable interacting with - which really does gets you far. That's really interesting from a social perspective, but computationally how effective is it?

Maybe I'm bike shedding, but I'm curious. I'm plotting the formulas via desmos.com to get an intuition on the behavior in a WOT.

Here we have the Coracle formula in orange where we assume both follows and mutes are the same (x-log(x)^2). I checked Coracle and see that you have a lower bound of 0. So what this graph says to me is that in the base case, if you weight mute and follows equally, you always get a positive score.

The second curve in red is the formula x-log(ax)^2 where a=24. So what happens if you gain mutes at 24 times the rate of followers? Well, only if you are under 4 followers will you be under 0. Now my judgement here is that the score here is helpful to the user socially, but functionally doesn't do much in terms of feed filtering.

A formula that is inverted around zero is probably the best way start off with because you have an easy to derive cutoff point.

Here, we have in green x^3 where X is some weighted sum of positive and negative interactions, which in the simplest case for a WOT would be something like (x-y)^3. If both mutes and follows grow at the same rate, you're stuck at zero, which kind of makes sense even if very lenient. However, you're in the negative the moment mutes grow faster than follows (blue curve, a=1.1) Considering that I'd expect the rate of gaining mutes to be much slower than follows, the next step I'd probably experiment with real data to see what kind of value for 'a' would be effective for feed control.