Placing a hot object on a 100% emissive surface shows a reflection where there shouldn't be one if that surface is defined with a ClassicEmissivity property (in contrast to a Ward BRDF).
Turns out there were two things going on here which is why we didn’t catch it:
1) there was a straightforward bug in how the path weighting was being applied to emissive components (now fixed)
2) the Ward reflectance model (what we use for most of our testing since it is easier to quickly setup versus class emissivity) was being “smart” when its emissivity was 1 and didn’t bother doing any sampling — the resulting ray never hit anything and therefore it appeared to have the correct behavior
ClassicEmissivity triggers the “pure” Lambertian optical property which “dumbly” does the same thing no matter the magnitude -- in this case sampling the background and making the bug in 1) visible
The fix will be in all following releases.