The scene is designed and physically assembled by scientists from Cornell University. They have measured the geometry of this scene and provided precise spectral characteristics of light source as well as material reflectance. All this data together with the reference image in High Dynamic Range (HDR) format is available here. We have converted spectra to rgb using the procedure described here and prepared corresponding 3DS MAX scene. Reference image (below left) is obtained from the reference HDR image using Gamma=1 (which is the same as using linear tone mapping curve) and brightness of +4.5 f-stops (anyone can repeat that using HDR shop program). Right image below is computed by BusyRay with linear tone mapping curve and MaxHDR set to 0.0035. Note that reflection from colored walls is slightly brighter on the BusyRay image. This is because of RGB approximation versus spectral one. But otherwise one can see that the match is quite close.

Cornell Box. Reference solution (left) and image computed by BusyRay (right).

It is known that in homogeneous environments (geometry is arbitrary, but all surfaces have the same diffuse reflection and emission) one can easily compute lighting analytically. Scientific reference is:

P. Shirley. Time complexity of Monte-Carlo radiosity. Proceedings of Eurographics '91 conference, pages 459-466. Elsevier Science Publishers, 1991.

Our analytic test is a closed box of constant reflectivity with constant self-illumination all over the surface. Direct illumination consists of only self-illumination which is set to 64 - this can be checked by rendering the box from the outside (in this case indirect illumination is absent). After each indirect bounce of light inside of box, light energy is decreased by diffuse albedo denoted below as Dr. It is easy to see that after infinite number of bounces luminance can be computed as infinite sum:

We have set diffuse albedo to 128/255. According to formula above, luminance of the box from the inside must be equal to 128.5. BusyRay 1.3.11 gives out slightly less because we compute limited number of indirect bounces at the moment. This will be fixed in the nearest versions to come.

Analytic test. Gray box with constant self-illumination as seen from the inside (it must be constant gray with intensity as computed above)