Strange buckets in VrayMatteShadow Pass

[PIXELBOX_SRO]

2 or 3 is fine for some cases: tiny, or moving geo may need 9 rays to be found inside a pixel.
10 subdivs will converge the shadow/matte shadow passes when there isn't enough info in the RGB to go on by (the scene above has non-gammaed pixel values of ~0.01f. An auto-exposure would boost that image exposure by at least a factor of ten, and then adaptive routines would work in the shadow pass too.).
REs which make sense (and are quick enough) to be taken into account for AA have their corresponding checkbox.
If V-Ray took every RE into account for AA the render times would explode: the behaviour is identical to competitors, by the way.
A min AA of 10 is indeed a fixed sampling "padding".
Which is how data *passes* have historically been rendered: fixed AA for render elements where contrast is not nearly enough to have an adaptive algorithm do its job.
Nothing has changed in this as far as i recall.
Alas, mattes and masks were *always* relegated to special setups, *never* calculated with the beauty, on the jobs i witnessed and directed technically.
Speed and iterability were always worth the extra prepping effort (which, by the way, could be fairly easily scripted.).

Thanks a lot Lele, never though of it that way (applyin AA to passes multiplying rendertime)