Remember that one of the major breakthroughs of the World Wide Web was that HTML meant documents were no longer plaintext.
I like text-based formats, but I'm not convinced that "Being non-binary is a huge plus" for parsing. With binary formats you can assume that documents are generated by a tool, which is at least trying to be compliant with a spec, so barfing on noncompliance is more acceptable. With text you have to be prepared to cope with any kind of rat dance imaginable.
SGML has the SHORTREF feature which allows custom Wiki syntaxes such as markdown, but also casual math. It works by applying a context-dependent (parent element dependent) mapping of tokens (such as the `_` token for markdown emphasis) to replacement text (eg. the `<b>` start-element tag). Within the `<b>` context, the `_` token is mapped to the `</b>` end-element tag, in turn, ending the emphasis. In combination with tag omission/inference (such as in HTML) and other markup minimization and processing features, SGML is a quite powerful plain text document authoring format.
[1]: https://en.wikipedia.org/wiki/Standard_Generalized_Markup_La...