So instead knowing city -> offset mapping, you need to know both city -> TZ and TZ -> offset mapping, and end up with something worse (because TZ will never be

So instead knowing city -> offset mapping, you need to know both city -> TZ and TZ -> offset mapping, and end up with something worse (because TZ will never be perfect, so offset from TZ is usually worse than "human" offset based on the location). I really see no benefit.