: Reach's rain works by spawning "raindrop" particles into the game's
: particle buffer; to some degree, they exist as "real" objects in
: world-space. The raindrop-type particles are set up to create splashes
: when they collide with something.

: Reach's GPU-driven particle system checks for collisions by comparing a
: particle's location to the z-buffer; when collisions happen, it can
: reflect the particle based on the surface normal at that location (and
: taking an elasticity value into account). If a particle's velocity dotted
: with the normal of a surface is positive (i.e. if the particle is
: approaching from behind the surface), the game's particle system
: supposedly (according to Reach's effects tech presentation) deals with
: this by removing the particle for being stupid, rather than let it do
: bizarre crap at the backface of a surface.
: This is why I'm surprised, and might take a closer look. It would be one
: thing if the raindrops were falling through when I'm not directly looking
: at the platform's underside; they wouldn't experience collision if there's
: nothing in screen-space to collide with! But when the surface is in view,
: they shouldn't be falling through it. But what if the game is spawning
: the raindrops very close-by (i.e. on and around the underside of the
: platform).

: It would be interesting to see the logic it uses to spawn raindrops. :)

Maybe the rain region is is being sliced so there's a new region underneath... it would useful to know if the rain is actually splattering on the underside of the surface texture, or if it were hitting an invisible barrier just below the texture.
Ideally, we should observe if the rain drop existed above the texture and continues to the expected destination below the texture (to determine if it's a new particle below or the same particle passing through), as well as where the 'splash' itself is being created (to determine the cause for the splash detection - collision with the texture but on the wrong side, or with a misaligned barrier of some sort).
Might be as simple as the surface normal being incorrectly inverted?
I have not looked at any of the Reach graphics tech specs (obviously :), just pondering.