STEMKit

Browser tools for computational chemistry, plus @stemkit/core, the tested library underneath them.

All computation is client-side. No uploads, no accounts, no install. Tools work offline once the page loads.

Live: https://stemkit.net

Quick start

Serve the site (static, no build step):

git clone https://github.com/LD-Shell/stemkit.git
cd stemkit
python3 -m http.server 8000     # then open http://localhost:8000/

Opening the HTML files over file:// will not work. Fourteen tools load ES modules, which browsers block outside HTTP.

Use the library:

npm install
npm test                        # 1077 tests, 16 modules
node tests/smoke.mjs            # end-to-end against a real install

import { parseXvg, columnStats } from '@stemkit/core';

const { matrix, headers } = parseXvg(readFileSync('rmsd.xvg', 'utf8'));
console.log(columnStats(matrix.map(r => r[1])));

npm scripts

Script	Does
npm test	Jest, 1077 tests
npm run test:coverage	coverage report (see docs/COVERAGE.md)
npm run check:links	internal and external link check
npm run check:links:internal	internal only, no network
npm run build:css	rebuild Tailwind after editing src/tailwind/input.css
npm run watch:css	same, on change

What is here

18 research tools:

Area	Tools
Data and statistics	plot digitiser, data cleaner, statistics calculator, error-bar generator, outlier detector, curve fitter, plot builder
Molecular simulation	XVG visualiser, structure inspector, coordinate manipulator, MD workflow generator (GROMACS, LAMMPS, PLUMED)
Writing and citations	BibTeX sanitiser, BibTeX deduplicator, DOI to BibTeX, journal abbreviator (ISO 4), LaTeX table builder, equation editor
Units	scientific converter: energy, length, pressure, dipole, polarizability, spectroscopy, temperature

Three further pages are workflow helpers, not research tools, and are not part of the scholarly contribution: Pomodoro timer, decision matrix, kinetics sandbox.

@stemkit/core holds the computation: 16 DOM-free domain modules, plus an aggregate export (index.js) and a dependency-injection layer (vendor.js). API reference in src/core/README.md.

Why the computation is a separate library

Client-side tools are good for privacy and bad for reproducibility: a figure produced by clicking is hard to regenerate six months later. Moving the computation into an importable library makes the same code path scriptable, version-pinnable and testable.

It also surfaced defects that had shipped:

• wavelength conversion returned a plausible but wrong number
• skewness used the sample, not population, standard deviation
• virtual sites in water inflated system mass by 67%
• adjusted G1 fed into the D'Agostino-Pearson transform, 21% error in K2 at n = 10

Each now has a regression test. Full list in CHANGELOG.md.

Docs

File	Covers
src/core/README.md	library API
docs/SETUP.md	layout, deployment, gotchas
docs/COVERAGE.md	reading the coverage report
docs/CSS.md	stylesheets and where rules belong
CHANGELOG.md	changes, including output-affecting fixes
CONTRIBUTING.md	contributing, and where code belongs
THIRD_PARTY_LICENSES.md	vendored library licences
paper/	manuscript, LaTeX and Markdown

Citing

10.5281/zenodo.21543112

Resolves to the current release. Machine-readable metadata in CITATION.cff.

Licence

MIT, see LICENSE. Vendored libraries under js/dependencies/ keep their own licences: THIRD_PARTY_LICENSES.md.