Own engine, no contenteditable
Document model, line breaking, pagination, caret, selection and DOCX reader/writer written from scratch. No ProseMirror, no editing library; the engine's only dependency is `fflate`. Whatever the editor doesn't understand is kept untouched in the file.
Measured against Word, line by line
Pages are laid out by Word's rules, and a command compares every test document against the PDF Word produces. Every layout change is validated with numbers, not by eye.
One engine, any runtime
`@rebodocs/core` never touches the DOM: it runs in the browser, Node, Deno, Bun and Workers. The same layout produces the PDF on the server without a browser, and templates are filled from an API or the command line.