What people who haven’t spent significant time trying to design naturalistic plain text interpretation rules don’t realize is that it is essentially a game of Whac-A-Mole – you change a rule to fix the behavior for one use case you end up breaking the behavior for other use cases. Only AI (or a pseudo-AI decision tree with thousands up thousands of branches) would come close to solving that.
As I said, if you believe that Markdown, CommonMark, reStructuredText and AsciiDoc are wrong (they all interpret 5. Klasse in der Schule as a list item), then you should design and propose a new plain text format, not try to change a format that is committed to remaining backward compatible to a huge corpus of existing documents that rely on the existing behavior.