Queries a benchmark's image matchups — a page of compared image pairs with how the responses split between them.

An item is one image pair as one leaderboard saw it, so a pair compared on several leaderboards comes back once per leaderboard and the boards' verdicts on it can be read side by side. Ordering (extremeness by default) ranks the matchups themselves — extremeness being how far the split sits from a coin flip, in standard deviations — where the prompt-ratings query ranks prompts by their most decisive matchup. Filters use that query's field set, applied to the votes a matchup is tallied from, plus WinnerParticipantId and LooserParticipantId, which select matchups by which side won the whole matchup; those two are only accepted as top-level and leaves, and a tied matchup matches neither. Matchups of leaderboards excluded from the benchmark overall are never returned. sides holds exactly two images, the one that took more responses first, and their responses sum to the matchup's. extremeness and the winning side are always read off the raw counts; useWeightedScoring changes only each side's wins.