It's the Arduino forums. It's for artists, by artists (by their own definition).
The test is completely meaningless as there is no picture of the layout nor schematic.
The author
seems to
almost understand that the limiting factor is not the IC but the layout and other circuitry, by saying: "Maybe 16 [bits] if you're careful with grounding and shielding and use a very low noise power supply and sensor." But accepting the consequence of this understanding, that the test is completely meaningless, is left for the reader.
But hey, when working with 24-bit ADC's, you
have to be careful with grounding and shielding and use low noise power supply and [low noise] sensor!
If not, the ADC magically cannot clean up after your mess; you are just measuring the noise you put in. Actually fairly benchmarking 24-bit ADCs would require quite some expensive gear and massive knowledge.
And, if the end user doesn't have such gear and knowledge, the results is limited to around 13-14, maybe 15 bits maximum
regardless of the IC.
The arduino forums can sometimes be a good source of information, but there is a nonobvious, cultural gap in the nomenclature. For example, when they talk about an IC using its specific name, they are almost always talking about the combination of the IC, a poorly designed module (connected using coat hanger like air wires), and a poorly written software library for it. You
can apply the information if you are going to use the same module and software library, but you need to know this.