openai/gpt-chat-latest → A (9:1)

Side A makes substantial architectural improvements: it removes the temporary demo-counter feature, introduces a dedicated settlement worker that batches vote persistence and ranking recomputation, adds cached ranking access (`ranked_items_cached`), warms the cache at startup, and switches UI reads from write locks to read locks. Side B is a focused UI enhancement that changes row coloring to use per-group score normalization with good supporting tests, but it is limited to presentation rather than core project behavior or performance.