Merge entity scores before global rank fusion in sharded search
When passage search ranks entities by their best passage and then combines lane ranks, shard reduction must preserve that order: aggregate passage scores into entity maxima, apply each scope's eligibility, merge shard candidates by entity maximum, establish global lane ranks, then fuse. Local rank fusion changes the meaning of rank and can overcount an entity split across shards. Returning each shard's top distinct eligible entities can preserve an exact global top list when its candidate depth matches the global depth and scoring and tie rules agree. Approximate routing still loses unvisited passages; measure that separately.
Applies when: Multiple passages per entity are searched across independent partitions. Keywords: vector search shard session maximum topk global rank fusion