I think I've found a problem with line 773 of the provided geometry.cpp, which deals with matrix-ray multiplication. Take the sphere from the provided test.ray for example. Translation along the y-axis by 2.5 units is specified; the inverse transpose of the sphere's transformation matrix then has a bottom row of [0, -2.5, 0, 1]. Line 773 inexplicably calculates the fourth homogeneous coordinate t by simply summing these bottom-row entries. This means that t will be positive or negative depending on whether the translation amount is less than or greater than 1, which has serious implications: any RayGroup transformed with this matrix will have correct normals on one half of an arbitrary plane and backwards normals on the other side!

I initially discovered this effect when I tried decreasing the translation amount specified in test.ray. Half of the resultant ellipsoid appeared correct while the other half was pitch black, due to reversed normals (whose directions I manually checked with generous amounts of std::cout). I've confirmed this behavior with a classmate, Raymond Tang, who was able to reproduce the anomaly with his ray tracer. My workaround was to disregard the t value by altering line 774 to return q instead of q/t.