LaTeX formulae from English Wikipedia

This dataset is the reference of EnWikiFormulaeTest in the Math extension. Both files map the MD5 hash of an input from the English Wikipedia to a string:

• en-wiki-formulae-good.json: the 672,902 inputs that WikiTexVC accepts.
• normalized.json: the TeX that WikiTexVC returns for each of them.

The inputs of all Wikimedia projects are in 10.5281/zenodo.15162181.

The first version appeared as test/en-wiki-formulae.json in texvcjs.

Format

The input is the content of the tag as the Math extension receives it:

• <chem> and <ce> tags are wrapped in \ce{…}.
• The attribute display="block" wraps the input in {\displaystyle …}, display="inline" in {\textstyle …} and display="linebreak" in \[ … \].

Pipeline

The files come from the dump of 2026-09-01 (10.5281/zenodo.23098003), created with the pipeline described in the README of wikiFilter and the Math extension at commit 9e3bd130 of master. Every run is recorded in the log.

Changes

• The inputs come from the dump of 2026-09-01 instead of the snapshot of 2014. 140,187 of the 312,190 inputs of version 1.0.1 are still used, and their output is the same, except for \Omicron, which WikiTexVC now writes as \mathrm {O}.
• Inputs with a display attribute keep their wrapper (see Format), so their hashes differ from earlier versions.

Limitations

• A math tag without a closing tag takes the rest of the page as its input, as in MediaWiki itself.
• Formulae that come from templates, or that are written without tags, for example with the {{math}} template, are missing.