Thanks grin

I flip in the first render pass the figurine along the mirror plane and write the UV of the original pose in a RT... then I render the scene, writing out a mask for the mirror mesh and the approximated refraction vector from the difference between the normal and the mirror plane normal. In a PP combine pass, I lookup for each pixel of the mask the UV of where to lookup SSR.. then I apply an additional fade-out value, which I computed from a fresnel term (from the 1st pass) and the re-projected viewport borders + a fadeout on the distance to the mirror plane.

It's not as versatile as SSR, but it looks pixel perfect on polished planes and that is what I want laugh I am happy!