That's a nice thing about the way that Attackpoint deals with GPS data: it ignores the bogus altitude from the watch, and just uses your coordinates to look up the altitude on a USGS map. So the track that shows up on your log has the altitude behaving nicely, never going below +38 meters (where you turned onto Russell St.).
You'll have to check with Kenny, but I think he uses at least 3 sources for altitude data, with a hierarchy depending on their availability and accuracy.
Though, at least for the graph of the elevation profile
at the top of the page, it still doesn't realize that I take the bridge over the Connecticut River. :-)
Right. That's a challenging issue, the elevation models have only one altitude value for any X-Y point, and don't have a way to take into account the idea of a bridge. I'm not sure what the thinking is, but you might imagine that the decision was made to leave off buildings and instead show the elevation of the ground (because if you included buildings, would you then need to also include certain types of vegetation?). And then maybe bridges get categorized as buildings and left off. It's a problem without a perfect solution, so they made reasonable compromises.