Specify HistFactory v2 Gaussian-Process Hyperparameters

Determine whether the HistFactory v2 specification should prescribe Gaussian-process hyperparameters, including the kernel choice and interpolation code, or leave those choices to individual implementations.

Background

HistFactory v2 replaces the traditional factorized interpolation assumptions with Gaussian-process regression, allowing joint systematic variations and providing interpolation-uncertainty estimates. Its serialization format must support multi-parameter variations and optional gradient information while remaining compatible with the HST standard.

A design question remains concerning the degree of prescription imposed by the specification. If kernel choices and interpolation implementations are standardized, models may be more reproducible and interoperable; if they are left to implementations, frameworks may retain greater flexibility. The paper explicitly identifies this choice as an open question.

References

Open questions raised at the workshop include whether the specification should prescribe the GP hyperparameters (kernel choice, interpolation code) or leave these to the implementation.

— IRIS-HEP 2026 Statistical Ecosystem Blueprint White Paper  (2609.19992 - Feickert et al., 17 Sep 2026) in Section 6, “HistFactory v2”