See this comment from the KHR_materials_volume PR:
The LOD we are using to read from the refraction texture is now also taking into account the volume IOR (in addition to taking into account the roughness/alphaG):
float refractionAlphaG = mix(alphaG, 0.0, clamp(vRefractionInfos.y * 3.0 - 2.0, 0.0, 1.0));
float refractionLOD = getLodFromAlphaG(vRefractionMicrosurfaceInfos.x, refractionAlphaG);
vRefractionInfos.y = 1 / volume_IOR.
Given the formula, if the volume IOR = 1 (which is the case with the file in the sandbox),
refractionAlphaG = 0 (whatever
refractionLOD = 0.
It seems the formula should be changed so that
alphaG still factors in even when volume IOR is 0…