This is a really nice analysis!

And I only, to my recollection, made the nut-murderer comment to counter what I thought was an overly strong position, so I agree mostly with James that Silvercloud's refinement is a winning argument if you have to rank-order intent and impact. (I do not agree that the typical state of affairs is that you're forced to rank-order them--interactions are usually more nuanced than that.)

Many times, impact matters more than intent. But the same nut-murder scenario I gave the first time illustrates that even a flat "impact matters more than intent" statement can also be wrong, because the ongoing interactions (for which intent is usually the more salient feature) can end up ultimately having an even greater impact.