Best within-scope ranking query

Determine the best ranking query for ordering candidates within the relation-localized scope of NativeScope, including whether the target term alone or the full question provides superior retrieval performance.

Background

NativeScope and NS-FullQ use the same anchor, native relation scope, candidate chunks, encoder, and token budget, but differ in their within-scope ranking input: NativeScope ranks with the target term B, whereas NS-FullQ ranks with the full question q. The experiments find only small and statistically inconclusive differences between the two variants. Consequently, the paper identifies the optimal within-scope ranking query as unresolved rather than establishing that either formulation is generally superior.