Before I move on to Fresnel and other things, I’d like to know if this is correct.
D diffuse
S specular
L lambert diffuse (max(0, n dot l) * 0.5 * D) + 0.5 * D
C cubemap color from reflected vector
F final color (L + (C * S)) / 2
Figured it out. Instead of the weight being 0.5 for both L and (C * S), it should be the brightness (the max rgb value) for the specular term and one minus that for the diffuse term.
Whoops. The max weight is only on the diffuse. The specular is just (C * S).