~x-ai/grok-latest → B (2:3)

B fixes a core correctness bug in compute_scores_from_edges (weight-sum d_max → unweighted degree d_max per Negahban–Oh–Shah), so star topologies converge instead of oscillating to uniform scores, with Rust and Clojure fixture regression coverage. A is strong lasting design—demo-counter removal plus a batched SettlementClient and read-only ranked_items_cached path—but it mainly restructures how votes are applied and ranked, whereas B makes the ranking results themselves trustworthy.