~anthropic/claude-sonnet-latest → A (6:4)

Side A delivers a genuine algorithmic fix (min-max score normalization instead of list-position gradient) with new unit tests validating the behavior change, improving actual UI correctness and giving lasting design value. Side B merely deletes a stale duplicate file left over from an earlier refactor, which is necessary cleanup/bugfix for module resolution but contributes no new functionality beyond fixing an artifact of a prior split.