Why colour detection sometimes seems inexact ?

Colour detection is complex matter. As it is done using the device's camera input, it highly depends on the camera sensors quality (which is generally very good on Apple Devices) but also on the lighting conditions. Cameras are generally calibrated for daylight lighting environment use (ie outside under a normal sunny day). This means that getting colours on a different environment can lead to very different results and this is expected. For instance, if you try to detect colours inside of your house lighted with incandescent or neon lights you can get very different results.

To achieve best results, try to detect colours under outside lighting : Go outside or detect close to a window.

Using the torch is also a good idea under low light conditions. However, the object you point to needs to be mat so that it does not detect the torch spot that may appear on glossy material such as glass, lacquer painted objects...  

Why Colours? does not use well known colour names such as Ultramarine blue, Sand, Red wine, Lemon yellow,...?    

Simply because these names are not colours! Let me explain this a bit more in details. First, each person perceive and feel colours differently so it is hard to have a common colour reference other than basic colours. This is already complex enough for people to agree about violet and purple values, to tell if a blue-green is more green or blue... Second, these colour names were often given by paint manufacturers to name their paints and some have reached the common language with, at best, an approximative colour definition. For instance the colour of the sand from the Sahara is very different from the one in the Mojave desert, the colour of a Bordeaux red wine is very different from a Bourgogne.

To name colours, we had to stick with a common colour standard. The only one that is accurate enough in terms of colour value without being too vague in its naming is the NBS Dictionary of colors. That's the reason why we use it for naming our colours.