About Automated Product Inspection

This is not limited to "image inspection.
I believe that every "automatic inspection machine" on the market today extracts "information" from the object to be inspected in some way, and makes a judgment as to whether or not the "information" is what was expected.
The problem here is
Can we represent "all of the inspected objects" as a finite number of pieces of "information"?
Even if it is possible, it may be very difficult to determine "from where", "how", and "what" information to obtain.
How can we determine the expected value of such "information"? What is the basis of the expected value?
How can we determine the expected value of the "information" and what is the basis of the expected value?
Can we introduce high-precision measuring equipment, measure 1,000 items such as dimensions, etc., and say that the product is a good product because the 1,000 items were in line with the expected values? It may be that we are trying to do something quite reckless.
In the end, I feel that there is no way for an automated product inspection to define a "good product" by a person and inspect it for any differences from a good product by as many means as possible. As long as the product is the same, the result should be the same no matter what means are used. If it is different, then it is different from a good product, and that is all.
The trouble here is the "variation of good products. There is no such thing as an absolute good product. Therefore, a "statistical range of good products" is obtained from a large number of "good products. Then, we inspect whether the product is within that range. As a result, it is not a question of whether the expected values are met, but how well they are aligned.
This is the FlexInspector methodology, and even if we remove the "image" framework, it is still an unusual way of thinking. However, if you consider "good quality = small variation in products", it is not such a strange way of thinking.