(...Okay but actually if you take the perspective that Aqqord's engine is going to behave like a not-very-debugged machine learning system, it's perfectly natural for the top results to be weird. Consider Opalyn's real value function in mates to be V, and view Aqqord's machine as computing a proxy U of V, with an error term [U - V]. Indeed, view U as being composed of V plus the error term [U - V]. Then selecting the topmost results from U = V + [U - V] is not just selecting on high V (insofar as U correlates with V at all), but is also potentially selecting for places where the error term [U - V] is positive and high. This is the generalized form of Goodhart's Law that we might combine with the Optimizer's Curse to yield a concept of Goodhart's Curse; to select high values from an imperfect proxy is to select for upward error in that proxy, even if the proxy is itself statistically unbiased on average.)
(Or put less technically: Just like search engines in Opalyn's world from time to time start to cough up spam or other weird things in their top results, it shouldn't be surprising if the top results from Aqqord's engine are a collection of weird cases that sideways maxed out Aqqord's imperfect attempt at a search algorithm. Like result #2 being that one guy in town who everyone dates, or result #3 being some overly-determined fellow with the capacity to make any relationship work if he tries it at all, or result #1 being the all-around objectively best person. It shouldn't be that surprising if you have to go down to #5 to start finding interesting people who might actually match Opalyn.)