Fully data-adaptive calibration of e-processes and variance proxies

Develop fully data-adaptive calibration procedures for the underlying e-processes and tuning parameters, particularly the variance-proxy parameter, while preserving anytime-valid FDR control and sample-complexity guarantees.

Background

The theoretical guarantees depend on the quality of the e-process construction and on tuning choices such as the variance proxy used by e-PS. The paper explicitly identifies the development of fully data-adaptive calibration procedures as unresolved. Such procedures are intended to reduce reliance on externally specified or potentially misspecified tuning choices, especially in human-facing applications.

References

Our theoretical guarantees also depend on the quality of the underlying e-process and on tuning choices such as the variance proxy parameter; developing fully data-adaptive calibration procedures is an important avenue for further study.

— Sample-Efficient Multiple Testing with Adaptive Data Collection  (2609.26651 - Lin et al., 22 Sep 2026) in Section 6, Discussion