openai/gpt-chat-latest → B (3:2)

Side B fixes a correctness bug in the core ranking algorithm by changing the Rank Centrality transition matrix to use degree-based d_max instead of summed edge weights, preventing oscillation on star topologies and producing the expected stationary distribution. It also adds targeted Rust and end-to-end regression tests with ranking fixtures, whereas Side A is largely a refactor that removes the demo counter and introduces cached ranking/settlement infrastructure without an equally clear correctness improvement.