Methodology

Source materials

Materials in this edition were generated from the compressed zip file available on the CD distribution page of the e-Handbook. The content of this file was compared against the e-Handbook’s online materials; discrepancies were few, and where they mattered they are called out in the text of this edition.

Code conventions

Code is intended to be pragmatic, readable, and — most importantly — reproducible. New code favors functions from the tidyverse over base R or the original Dataplot/S-PLUS scripts, which are retained alongside the original handbook content for reference.

Diagram conventions

Where possible, diagrams were recreated as SVG files (or as R/ggplot2 figures rendered directly from the chapter’s own code), so that they can be edited independently of any particular software program. Where it was not practical to reproduce an original figure faithfully this way, the original GIF image from the e-Handbook was retained instead.

Review methodology

Every worked numerical example in this edition was checked against the value reported in the original e-Handbook, generally by reproducing the calculation in R (and, where one exists, cross-checking against the e-Handbook’s own companion Dataplot or R script). Genuine discrepancies between the original text and the recomputed value are flagged in place with a callout note, rather than silently corrected — the goal is to preserve the original NIST content as the reference, while being transparent about anywhere this edition’s numbers diverge from it.

A full line-by-line comparison of the original archived materials against this edition was not feasible — the original e-Handbook runs to more than 1,300 individual pages. Instead, each section was built and spot-checked for accuracy as it was migrated, and whenever a deviation from the archived material was detected, the surrounding section received a careful, dedicated inspection. Undoubtedly, additional corrections will be needed as this edition is used and reviewed more broadly — see CONTRIBUTING.md in the repository for how to report them.

Statement on the use of generative AI technologies

This edition was produced in close collaboration with Claude Code, Anthropic’s agentic coding tool, working under the direction and review of the editor throughout.

For each chapter, the editor directed which section of the original e-Handbook to migrate next; Claude Code then read the corresponding source page(s) under handbook/, drafted the Quarto (.qmd) markdown and R code reproducing the worked examples, executed that R code to verify every reported numerical result against the original text, and rendered the chapter to confirm the page built correctly before moving on. Genuine discrepancies between NIST’s published values and the recomputed ones were flagged with callout notes rather than silently altered; the editor’s own transcription or formatting corrections were applied directly. Decisions about structure, tone, visual conventions, and how to handle ambiguous or inconsistent source material were made by the editor in the course of this collaboration, and captured as standing conventions for the rest of the project as they arose.

The editor reviewed each section as it was produced, spot-checked calculations and rendering, and made the final call on anything flagged as a discrepancy, an ambiguous judgment call, or a structural decision. As with any large-scale migration of this kind, responsibility for the accuracy of the final published content rests with the editor, not with the tool used to help produce it.