Nobody cares about the value of middle grey. It’s just a math anchor to unroll image processing algos. There is no truth in it. Normalized small dynamic range RGB spaces place it between 18 and 20% luminance, display-referred. Normalized spaces are useful when dealing with non-normalized input and output spaces, so they serve as connection spaces.
Middle-grey is just a concept of an average luminance value, at equal distance between “white” (or max luminance) and “black” (or min luminance). And because, these days, “white” can mean “diffuse white patch at 20% reflectance” or “specular highlights in HDR”, middle-grey is a better anchor than white.
No. Your main object should have the luminance value you want it to have. But in well-exposed image, that luminance often falls close to standard middle grey. So, you might use that property to setup your exposure a priori in case you are lost. But it is a posteriori statistical description, not a an image-processing prescription.
Does it look good ?
- if yes, keep it this way
- if no, change it.
You should really relax with numbers (says the guy doing maths).
Except if you have several light sources (and every image is at least direct light + bounced light), you have several “whites” and several “middle-greys”, so you still need to choose which one is your exposure anchor.