EXPERIMENT VIEW
Compare 32, 64, 128, and all 256 candidates
For every ground-truth atomic review issue, this page counts how many candidates have a frozen GPT-5.6 Sol-high match score of at least 0.70. Switch between the full 256-candidate pool and one seeded-random representative from each of 128, 64, or 32 spherical-bisection clusters. Counts are independent per issue; no Hungarian one-to-one restriction is applied.
Loading ground-truth match counts…