Misinformation / Disinformation

Two sentences describing the same event:

"Protesters gathered outside the building.""A mob assembled outside the building."

Same event, same verifiable facts, entirely different reader. No false statement in either.

This is the layer of a text that survives fact-checking untouched, because the manipulation is not in what is claimed but in what is connoted. Word choice does work that the argument never has to do explicitly.

The exercise that made this concrete for me: rewrite a paragraph replacing every loaded term with its neutral equivalent, then read both. Often the argument gets noticeably weaker — which tells you how much of it was carried by vocabulary.

That "so what remains?" test is the only reliable one I have found, and it does not scale.

For anyone building tooling in this space: is connotation something you have attempted to detect, or is it too dependent on context to be worth attempting?