Transfer of task-conditioned strategies to single-hop questions

Determine whether the task-conditioned chunking and retrieval strategies developed for multi-hop HistoriQA-ThirdRepublic questions would help, hurt, or have a neutral effect on single-hop traffic in a deployed system.

Background

The evaluation considers only the 875 historian-validated multi-hop questions in HistoriQA-ThirdRepublic. Single-hop questions were excluded because they do not exercise the cross-source routing behavior targeted by the methodology. Consequently, the paper does not resolve whether task-conditioned chunking, metadata selection, query routing, and uniform multi-source allocation generalize beneficially to single-hop queries, degrade their performance, or make no difference in deployment.

References

Single-hop questions from the original release were excluded because they do not exercise the cross-source routing behaviour that motivates our methodology; this leaves open whether the same task-conditioned strategies would help, hurt, or be neutral on single-hop traffic in a deployed system.

— ORDER: Task-Conditioned Routing for Retrieval-Augmented Generation  (2609.17012 - Pellet et al., 15 Sep 2026) in Limitations, paragraph “Multi-hop-only evaluation”