How does the built in shadow mapping work?

I'm trying to understand the built in shadow mapping system in order to fix a bug with my code. I am trying to project a screenspace space grid onto a worldspace plane (in order to sample a height map uniformly in screen space to implement LOD).

I have a bug where the shadows are not being drawn at the right places. I did some experiments with modifying the vertex coordinates in the ShadowCollector pass and it seems like the transforms being done on the vertices are in light space rather than camera space. Is this true? Are the built in transform matrices (m, v, p) set to different values than in the normal shader passes?