Hydrological modeling gets tricky when water flows underground, either naturally or unnaturally. A pixel surrounded entirely by higher elevation pixels is a "sink", and if you want to model the overland flow of water a preprocessing step is often to fill sinks and apply a slight gradient to them.

I believe that the NHD+ dataset uses an alternate strategy; it burns an artificial channel into a sink, connecting it to a lower-elevation outlet. The dataset includes attributes that identify these "underground flow" segments. I didn't examine ESRI's Living Atlas site closely enough to see if they've used a different symbology to represent these segments, but your comment makes it sound like they did not.

A pond often has no inlet and no outlet (as opposed to a lake), so it makes sense that some ponds may not show up in this dataset.