That is what Chrismuc also said. But, it is (quite) a bit more complex than that.
The colours from the sensor are not directly usable, but will be converted to normal RGB colours using some math, basically a matrix multiplication.
On top of that matrix multiplication there will one or more profiles will be applied and those profiles may have significant tuning for pleasantness or "memory colours". The profiles may be matrix- or LUT-based but often a combination of both.
So, the colour rendition is decided by a series of matrix multiplications, the first one (colour conversion matrix) is actually given by the sensor while the rest is visual adaption and tuning.
By the way, DxO presents the colour conversion matrices for each sensor. They also calculate something called System Metamerism Index, which essentially measures how well the sensor can reproduce the colour fields of ColorChecker card. That figure is not a part of the DxO rating, however.
Just to mention, the ColorChecker card is not just a standard used with 16 arbitrary chosen colour fields, but those fields are chosen with great care. The two skin patches, for instance, are said to show very similar spectral response to human skin. So those fields would be a good representation of human skin under almost any illuminant.