eval-harness

metric

pair_agreement

better
higher
kind
quality: measured against a reference or by a judge
decides
no goal
weighed by
no goal
recorded on
summeval_100

01what it measures, and how

Pairwise mode: among the sampled pairs of an article's summaries that the experts rank apart, the share where the judge preferred the summary the experts prefer; a tie counts half and an unreadable answer 0. Chance is 0.5.

A project explains its own metric at length with METRIC_DESCRIPTIONS = {"pair_agreement": {"summary": "…", "details": "…"}}.

02on summeval_100

gold/summeval_100se_gemma_l_n100_pairwise_v1: 0.825se_qwen_s_n100_pairwise_v1: 0.812se_random_n100_pairwise_v1: 0.4730.473better · 0.825

scored against silverscored by a judgewrote the reference it is scored againstscored against a label set, not a reference setunmarked rows are scored against gold