The definition of candela given by BIPM is as follows:
The candela, symbol cd, is the SI unit of luminous intensity in a given direction. It is defined by taking the fixed numerical value of the luminous efficacy of monochromatic radiation of frequency $540\times10^{12}\,\mathrm{Hz}$, $K_{\mathrm{cd}}$, to be $683$ when expressed in the unit $\mathrm{lm\,W^{–1}}$, which is equal to $\mathrm{cd\,sr\,W^{–1}}$, or $\mathrm{cd\,sr\,kg^{–1}\,m^{–2}\,s^3}$, where the kilogram, metre and second are defined in terms of $h$, $c$ and $\Delta \nu_{\mathrm{Cs}}$.
So, from this definition we can only directly characterize luminous intensity of 540 THz light. To find luminous intensity of light with any other spectrum we need the normalized luminous efficacy function (to be scaled by $K_{\mathrm{cd}}$). But this one is not so definite: there're at least three versions of it, shown below in black: CIE 1931 (solid), Judd-Vos modified (dashed), and Sharpe, Stockman, Jagla & Jägle 2005 (dotted).
(source)
So which of these, if any, does BIPM reference with its definition of candela?