~x-ai/grok-latest → A (5:1)

A lands foundational pairwise-voting design—new pair selection (bridge-preferring components), a full /vote compare UI with in-place morph after record_vote, ItemId::from_storage normalization, and solid integration/e2e coverage—whereas B only threads an optional pool param and a 'vote on children' link through existing garden vote helpers. A’s changes are larger in lasting product surface and algorithm quality, not merely line count; B is a useful but incremental constraint on already-present compare flow.