Principled coverage-based vocabulary selection

Derive a principled sample-selection rule for the misi vocabulary that replaces the heuristic coverage resampling direction and targets the per-query confusable count relevant to verification cost.

Background

Uniform sampling fixes the mean posting-list length but can produce severe load imbalance on hub-heavy data. The paper’s coverage-resampling heuristic improves recall at small verification budgets on some datasets but can become neutral or harmful when the budget is large. The authors therefore identify the need for a principled selection criterion, with the measurable confusable count from the vote model suggested as a natural target.

References

The coverage heuristic needs a selection principle rather than a direction that happens to win; the per-query confusable count that Theorem~\ref{thm:vote} makes measurable is the natural target for calibrated stopping. These are stated as open because they are.

— misi: a Metric Inverted Sample Index  (2608.27422 - Chavez, 27 Aug 2026) in Section 6, “Limitations and future work”

Furthermore, extending pair coverage captures more confusions but also lengthens the inference prompt and increases cost, and we have not tested how this balance changes for multilingual inputs, substantially larger taxonomies, or temporal drift.

— From Confusion to Clarity: Confusion-Aware Retrieval and Knowledge Injection for Text Classification  (2609.01564 - Gupta et al., 1 Sep 2026) in Limitations section

More broadly, our formulation extends to generative retrieval and to learned sparse expansion, and the finite sample behaviour of the breadth-consistency trade-off remains open.

— Tracing and Taming Hallucinated Terms via Subsequence-Association Filtering and Breadth-Aware Selection for LLM Query Expansion  (2610.13810 - Jaiswal et al., 9 Oct 2026) in Section 6, Conclusion, p. 12