About this site
How this site was made, how it was checked, where its numbers come from, and how to cite it.
Made with AI assistance
I built this site with the help of an AI system, Claude, made by Anthropic. Claude wrote much of the code and the first drafts of many pages, following my direction. I chose what the site covers, decided how each thing should be explained, and checked the results; the scientific responsibility for every page is mine.
The research the site presents is not the AI's work. The measurements, analyses and evaluations credited to me, on the evaluation pages and in my publications, are my own, done with my collaborators and published under our names; all other work is credited to its authors where it appears.
This is also how I think tools for our field can be kept alive: with an AI doing much of the routine work, one researcher can build and maintain resources for the community (an atlas of reactions, evaluations anyone can rerun, a rate explorer) that would otherwise be hard to keep up. It does not replace checking, which is why the next section is the longest.
How it was checked
Automated tests check the site:
- the atlas (75 reactions, 353 references) is validated: every reaction must balance in charge and mass number, every link must point to something that exists, and every reference must record how it was checked;
- the reaction rates and the network calculations are compared with pynucastro, a widely used Python implementation of the same REACLIB rates, and every number in the curiosities is checked against the explorer's own solver;
- the evaluations that run in your browser are compared with native AZURE2 on every model, and the published fits are the native results;
- the site search is tested on the queries people type, in English and in Italian.
Every number in the atlas and on the explanatory pages carries a citation to its source, and during the reviews the references were checked against the sources they cite. If you find one that is wrong, the report links below are the quickest way to tell me.
Each round of changes was also read by separate AI reviewers, prompted to take the view of a specialist, an accessibility reviewer and a general reader. They are not people and no substitute for them.
Data, sources and licences
- Masses and half-lives: the Atomic Mass Evaluation AME2020 (M. Wang et al., Chin. Phys. C 45, 030003 (2021)) and NUBASE2020 (F. G. Kondev et al., Chin. Phys. C 45, 030001 (2021)). The atlas's Q-values and half-lives come from them; the rate explorer shows REACLIB's own Q-values, those its rates were fitted with.
- Reaction rates: the JINA REACLIB library, read through pynucastro (BSD 3-clause licence). Each rate keeps its REACLIB label and its source paper.
- R-matrix: AZURE2 (R. E. Azuma et al., Phys. Rev. C 81, 045805 (2010)), the code I maintain, runs every evaluation.
- The evaluations' data: the measured cross sections in each of the 8 evaluations come from the papers listed on its page, and the input files can be downloaded and run locally.
- The data explorer: measurements from EXFOR, the international database of experimental nuclear reaction data (IAEA/NRDC, CC BY 4.0), and nuclear levels from ENSDF, read from a dated snapshot and linked back to every entry.
Open code, errors and contributions
The source code of the whole site, including the evaluations and the rate explorer, is public on GitHub: github.com/skowrons94/Website.
If you find an error, please open an issue; every reaction in the atlas, the s-process page and each glossary entry also has a “Report an error” link that fills in the report for you, and a “Suggest a reaction” link for what is missing. You can also write to me by email. Corrections and contributions (a reaction, a reference, a translation) are welcome as pull requests.
How to cite this site
J. Skowronski, Nuclear astrophysics: reaction atlas, R-matrix evaluations and reaction rates (website and source code), version 1.0.0 (2026-10-13), github.com/skowrons94/Website.
A DOI will be added here when the first release is archived on Zenodo.
For one reaction or one evaluation, cite the page itself (each has its own box) and, above all, the original papers it draws on.
Machine-readable: CITATION.cff