There is something that concerns me about the very fundamentals of what you propose.
Refer to Page 5 of your PDF document, headed "Theory of Operation"
The ray tracing from emissive surface 'A" is just fine. Photons leaving surface "A" travel for a brief distance through air (or vacuum), and then strike the refractive layer, and are refracted at the interface. No problems.
The ray tracing from surface "B" may just be a bit dodgy, because you have apparently assumed that it is posssible for the photons to start their journey from just inside the refractive material. Whether that is valid is unclear. Consider the situation if the emissive surface "B" was simply pressed hard up against the refractive layer. In that case, there would still be a very small air gap, and your ray tracing would be wrong - in fact, the ray trace would look identical to the rays leaving emissive material "A", and the device would therefore not work. As I understand, what you actually do (in effect) is to "paint" the emissive surface onto the refractive surface. It could be argued that the photons still originate outside of the refractive material, and will therefore be refracted as they enter it, just as if there was an infinitesimally small airgap between the two. If this is true, your device will behave symmetrically, and will not work.
It is easy to verify experimentally that it is possible to optically bond an emissive surface to refractive material so that refraction does not occur between the emissive surface and the refractive material. If you lay a triangular prism on top of a sheet of printed text, it is possible to observe the total internal reflection effect--when attempting to view the text from certain angles, the text will disappear and be replaced by a reflection of whatever is on the other side of the prism from the viewer. In contrast, if you paint the surface of the prism, then you can view the painted surface from any angle you like, and it will never "disappear" due to total internal reflection.
The same principle is employed in fingerprint scanners. You have a triangular prism with a 90° angle and two 45° angles. You have a light source attached to one 45° surface and the sensor attached to the other 45° surface. When nothing is being scanned, total internal reflection causes the photons emitted from the light source to bounce off the inner surface of the prism and reflect to the sensor. But when a finger is placed on the surface, a temporary optical bond forms between the top of the fingerprint ridges. This optical bond enables photons to travel from inside the prism directly into the finger being scanned without being refracted
. As a result, the sensor sees an image of the light source where the ridges are not in contact with the prism, and the ridges where they are in contact with the prism. The brightness difference between the reflection of the light source compared to the reflection of the light source off the skin is used to determine which pixels are fingerprint, and which are background.
The diagrams below illustrate the concept; imagine the black circle to be a fingerprint ridge pressed against the prism surface and creating the optical bond. Instead of being reflected directly to the detector, the photons are mostly absorbed by the fingerprint ridge. If you have a triangular glass prism, it is easy to visually verify this for yourself experimentally.
As long as the emissive surface is optically bonded to the layer of refractive material (basically the difference between painting the surface of the prism directly instead of just laying it on a painted surface), the photons will not be refracted as they are emitted from the emissive surface into the refractive material.
Your experimentally measured temperature difference of 0.018K strikes me as very small, and does not support your theory beyond all possible doubt.
I regard the experiments as interesting, but not as ironclad beyond-a-shadow-of-a-doubt proof.