Does the assumption of a homogeneous pool of content and a corresponding homogeneous pool of readers affect this? Segmenting by topic and applying the algorithm per each pool might be necessary if one wants to have a "fair" assessment of quality.
We thought about modeling quality and agent preference with higher dimensional vectors, like in a recommender system. And to see if a user will upvote a specific content, you calculate the dot product. A simplified 1-dimensional user expectation and 1-dimensional quality can be seen as working on the scalars of the dot products itself.
There was a Master thesis which did some simulations with high-dimensional vectors and how many dimensions make sense. The result was that it basically doesn't matter.
In the end, we have to make a real-world experiment with the HN community to see if it works as expected.