the front-end could just check the account history.

This would be too late in the front end. The entire calculation would then have to be performed again and the order queried again... Hivemind is already the right choice, as the relevant score is already calculated there.

A custom_json operation would have the advantage that no additional API requests would be required, as certain custom_json operations are processed by Hivemind anyway. However, like you, I see the effort involved in securing this handling against "attacks" or only accepting it from a specific account (for example @steemitblog). Then there is the question of where/how Hivemind stores the data.
But quite feasible...

dedicated account

This leads me to the thought that there is/was something similar for beem. The account @fullnodeupdate contains (in its profile) data about RPC nodes and their performance. Unfortunately, these are no longer updated on Steem. However, this alternative also requires further checks to ensure reliability.

Ergo: Changing the score calculation with a static (code-based) list should not be difficult. However, the implementation of a dynamic list must be chosen wisely. We will have to discuss this further.