I appreciate you looking deeper into this and discussing it!
I'm wondering whether a mode might be a better approach? It seems like it would be more reflective of the general consensus, unless every witness decides to switch to a different number - which seems rather unlikely.
I did suggest a mode but another witness came up with a counter example where it would not work as expected. Say 18 witnesses each have different APR values and two of them have an APR of 0%, mode would result in 0% APR.