tamte

it's also pretty common to scale your scene (sources/colliders/etc) to m, run the sim and scale the output back to original size to avoid adjusting the forces or other settings if your sim setup is already tweaked to m

so it usually doesn't matter what scale the scene was built for since FX artist can usually adjust it (even differently for each sim if necessary)
of course it's ideal to be consistent

zf3d
I want to hear the official explanation of proportion, whether he is right or wrong

not official, but based on my experience

Houdini FLIP is not 100% physical and there are many things that can cause physically implausible result even if you keep the scale 1unit=1m

but generally it's about whatever you decide if you make sure you adjust values of your forces etc accordingly
incorrect scale is mostly visible by how slow/fast splashes fall under gravity so you just need to adjust it for your units
(similarly Surface Tension will probably also need to be adjusted accordingly )

for example:
if you say 1unit=1m then for simple sims you want to keep gravity -9.81 since it represents m/s2
if you say that 1unit=1cm then your gravity would need to be -981 (cm/s2)
etc.

(many other POP forces are a bit arbitrary and should be tweaked visually as the @density doesn't really translate to @mass and POP forces have Ignore Mass on by default anyway, so you would need to be aware of a lot if you wanted to rely on truly physical inputs)

jedmitchell
How would you accomplish the goal of matting this Karma render, with the accompanying alpha, over a plate?

render your beauty with the dome light hidden from camera and refractions (Contributions set to : -refract)
and then render pure refraction pass using UV map instead of plate or env map (whatever you want to replace)

then in comp use such refracted UV map to map any image you want and add beauty on top you will get much closer representation of properly refracted new plate in comp
of course to be done properly one would have to also consider refraction contribution per pixel etc, but it will be definitely closer than pure see through shader