[Re] Independent computational replication of two block-maxima applications with missing observations

[canglang-social]

Original article: Simpson, E. S., and Northrop, P. J. (2026), Accounting for Missing Data When Modelling Block Maxima. Environmetrics 37(2), e70075. https://doi.org/10.1002/env.70075

PDF URL: https://github.com/canglang-social/evmissing-empirical-replication/blob/main/paper/manuscript.pdf
Metadata URL: https://github.com/canglang-social/evmissing-empirical-replication/blob/main/metadata.yaml
Code URL: https://github.com/canglang-social/evmissing-empirical-replication

Scientific domain: Statistics (extreme value analysis)
Programming language: Python
Suggested editor: Editorial assignment requested; no preference.

This is a partial computational replication of the two empirical applications (Plymouth ozone and Brest surges), using an independently written implementation informed by the paper and inspected author-package source. The original simulation study is outside scope; no new methodology is claimed.

The primary comparison meets frozen printed-value tolerances for 113 of 120 scalars, including all 48 printed interval endpoints. Seven discrepancies are retained and separately explained with fixed reference-optimizer evidence. A single bounded uninterrupted upstream-retrieval acceptance preserved all primary flags and passed 96 solver checks. Existing dependencies were used; fresh installation and a second machine were not tested. Historical reference calculations were not rerun by that chain.

Supplement: https://github.com/canglang-social/evmissing-empirical-replication/blob/main/paper/supplement.pdf
Reproducibility guide: https://github.com/canglang-social/evmissing-empirical-replication/blob/main/paper/reproducibility-guide.md
Data access and published licensing basis: https://github.com/canglang-social/evmissing-empirical-replication/blob/main/DATA-PROVENANCE.md

Source data are obtained directly from version-pinned CRAN archives with SHA256 verification; no third-party raw data are mirrored in this repository. The package declarations and attribution are preserved. Code is GPL-3.0-or-later; author-created article material is CC BY 4.0, with explicit third-party scope. OpenAI Codex assistance is disclosed in the manuscript.

[canglang-social]

A provenance clarification for this submission: on 18 September 2026, the Renext maintainer stated that the package GPL does not formally cover its bundled sea-surge data and missing-observation intervals. ASNR permits scientific reuse, requests citation of Renext and ASNR, and specifies illustrative use only because the observations and processing methodology have since been updated.

The correction to DATA-PROVENANCE.md replaces our earlier reliance on the package declaration for these data with that scoped permission and limitation. Our Brest analysis reproduces the historical published example; its return levels should not be interpreted as current operational flood-risk estimates. No raw inputs are mirrored. Numerical results and the original v1.0.0 submission snapshot are unchanged.

The provenance correction is published in commit canglang-social/evmissing-empirical-replication@0c879ae. The private correspondence is retained as evidence and is not posted verbatim.