A typical matsim execution path is that an activity has a coordinate (either by itself, or via its facility), and from there needs to "bushwhack" (= non network walk) to a link.
We need some decision about the walk distance in that case. Should it be
the shortest distance (= orthogonal projection, or distance to node if outside range of orthogonal projection), or
the distance to the actual pickup point (end of link for qsim)
Right now, my intuition would be that the first is better, for the following reasons:
Can be computed in router, independent from implementation in mobsim.
Is physically more plausible. Why should we assume that someone walks 100km only because the link is very long, and the coordinate happens to be near its upstream end?
A disadvantage is that it might look visually weird, since these would show as very fast bushwhacking walks to the qsim pickup point:
The shortest distance to the link is short, resulting in a short bushwhacking time, but the visualization would move the agent in the same time and thus with much higher speed from (coord) to (pickup).
Our usual counter-argument would be, I suppose, that we do not really resolve beyond the link level, and if someone needs this look better, she needs to make shorter links.