Rank-Biased Overlap
Rank-Biased Overlap (RBO) measures how much two ranked result lists differ, without needing any relevance labels. Search golden sets are anything but lists it among approaches for measuring how thoroughly a new algorithm shuffles results — not good or bad, just quantity of change — and links the RBO paper at https://dl.acm.org/doi/10.1007/978-3-031-15037-1_11.
Role in Evaluation
It compares two orderings directly, so it works where Judgment Lists are sparse or biased. It tells you that results changed and by how much, not whether they got better; pair it with a quality signal such as side-by-side review, unit-test-style checks or Session-Based Evaluation. For another label-free ordering comparison see Kendall Rank Correlation.