Cone-gain normalization
= Cone-gain normalization
A diagonal model applying gains $g_j$ to cone signals. If a change transforms every relevant catch as $Q_j\mapsto k_jQ_j$, gains $g_j\mapsto g_j/k_j$ preserve the normalized outputs exactly. A general <illuminant> change need not produce the same channelwise factors for every surface, so exact cancellation is conditional.