[timini]

Found by rendering the data repository's per-study sample (rewire-bio/rewire-benchmark-data#196) through the disagreement view.

• The ordering rule pools the agreeing papers' comparisons, so one of them can rank the other model higher on its own. In 11 of the sample's 221 findings that happens, and the summary hid it: "4 papers rank MutPred above PrimateAI in 16 of their 20 comparisons" while one of the four ranks PrimateAI higher in 4 of 6, and the circles placed it beside the outlier. The summary now adds "Of those papers, ranks PrimateAI higher on its own (4 of 6)." It is only added when there are two or more agreeing papers.
• Summary sentences that start with a lowercase source name ("proteinbench primary benchmark evidence ranks ...") now start with a capital.

Tests in tests/disagreement.test.tsx cover both.