Yea, you need some highly abstract thinking to truly understand shadow mapping; in particular, the relation to the "camera" and "light" spaces. Once you do understand it, the code itself is not too difficult. Same applies to blurring. Soft shadows are generally implemented as shadow maps with variable blur width based on distance or occlusion criteria (though this is not the only way).

You can use PIX to inspect the intermediate render targets (such as the shadow depth map), if you find that your technique doesn't work in an expected way. Once you see the device state visually, it is much easier to determine what's wrong (if anything).