~anthropic/claude-sonnet-latest → A (8:2)

Side A introduces substantial, lasting core functionality (form template substitution, ranking/rank-centrality algorithm with regression tests, reducer state machine, vote comparison UI, UI action dispatch) that forms real application logic and infrastructure, even though the diff includes messy artifacts like shell-prompt-contaminated files. Side B adds only a single browser integration test for pool-scoped voting, which is useful for verification but far narrower in scope and does not add product capability or design.