Right you are, fixed.
For the topic at hand – driveway-sidewalk intersections – I’ve been arguing for zero features, but if a feature is mapped at all, it probably should be consistent with the double-feature representation of normal crosswalks.
Crosswalk modeling may seem like a glaring exception to the “One feature, one element” principle, but there are actually quite a few exceptions where we can optionally represent a single feature as multiple geometry types. Other examples include bridges tagged on roadways versus bridge areas versus bridge relations, highway ways versus highway areas, bus stop nodes versus stopping position vertices, and place points versus boundary relations.
In most of these cases, each geometry serves a different use case, and the simpler geometry type predates the more complex one and is kept for backwards compatibility. Maybe someday we can confidently de-tag the crossing nodes, knowing that routers and renderers can infer them when necessary. The one wrinkle is that tactile_paving=* means something different when applied to the crossing node versus the crossing way.