My lighting is done using klux (kilo-lux) values, so you can in fact measure lighting amounts outside, of the sun, inside rooms, to get actual klux values (klux instead of lux btw because lux values are so large that you get floating point trouble; kilolux fits floating point buffers quite nicely).

Pushing actual klux values into the lighting time-of-day system then should more easily give realistic lighting if energy is conserved as above. So I’d still recommend it for games, if only to make coming up with a more realistic result becomes easier this way.

The line in the shader is a fit of a linear function to the real conservation factor.

So in Wolfram, you get:

((n+8)/(8*pi)) vs (0.0397887357729839 * n + 0.3183) for n from 0 to 10

http://www.wolframalpha.com/input/?i=%28%28n%2B8%29%2F%288*pi%29%29+vs+%280.0397436+*+n+%2B+0.3183%29+for+n+from+0+to+100

This changes the offset at n=0 to exactly 1/PI and seems more correct for specular=0..100.

Remember that total energy out is spread over the entire hemisphere, so the intuition that a a certain value in gives the same value out is not generally correct.

((power+8)/25.13274)*2.793

So at a power of 1 it’s at the same level of diffuse lighting.

This way you can easily compute the final light using:

float diff = max(dot(N,L),0);
float spec = max(dot(N,H),0);
spec = spec*((power+8)/25.13274)*2.793;

float result = lightcolor*lerp(diff, spec, reflection);

It would seem to me this way the total energy always adds up to the original input color energy.