~x-ai/grok-latest → A (3:2)

A upgrades the core pair-selection algorithm (bridge sub-priorities for spanning-tree growth, then zip-order adjacent rank refinement with established-component tracking) and locks it in with targeted tests—ongoing quality of every suggested vote. B adds real product surface (pool query param, children button, threading pool through vote UI/JS), but it is largely parameter plumbing over the existing suggest_next_pair_in_pool path rather than deeper ranking design.