Resolve the unsettled effects of breadcrumb depth and query type

Determine whether the observed breadcrumb-depth dose-response and query-type interaction patterns for dense retrieval are reliable and generalizable across corpora, embedders, and query sets.

Background

The paper conducts exploratory analyses of how heading-path depth affects coverage-nDCG and how structure treatments perform across query types. The breadcrumb-depth analysis shows inconsistent or weak effects for the weaker embedder, while the query taxonomies differ between Wikipedia Featured Articles and QASPER, preventing direct replication of specific interactions.

Because these analyses were not among the preregistered confirmatory contrasts and display corpus- or model-dependent behavior, the authors explicitly leave them unresolved rather than treating them as established findings.

References

The analyses we left uncontrolled --- the depth dose-response and the query-type interactions --- are precisely the ones that remain unsettled.

— Does Document Structure Help Dense Retrieval? A Placebo-Controlled Ablation of Four Mechanisms Across Two Corpora  (2610.10170 - Kuehlkamp et al., 7 Oct 2026) in Section 6, “Conclusion”