It's oddly worded in the article - the limit is given by the inequality, above:
Bell's Inequality: in classical world, given three properties A, B, and C of any collection of countable things,
the number of them that possess A but not B, plus the number that possess B but not C, is equal to or greater than the number that possess A but not C.
That is a fact of the normal world. : that {A but not C} cannot be larger than the sum of the other two. {A & -B} + {B & -C} is therefore an upper limit on {A & -C}. That is the limit referred to.
As you can see, it holds in any collection of objects that either do or do not possess property B, communicate by cause and effect through space, and interact in ways that can be modeled in bi-valued logic (i.e. meaningful assertions of property possession are either true or false).
That limit is violated, in these quantum world measurements of entangled systems. {A & -C} is larger than {A & -B} + {B & -C}.
Dunno. Greater than the inequality allows, is the report.
3. The detection of starlight color is a local process, in the context could that not have influenced the correlation?
Partly dealt with in the last paragraph.
4. Is the claim of people involved in the project that since random generator based detector is just a few microseconds process while starlight was 500 light years earlier origin and hence we have pushed the matter by an order if 16, sensible? Is it not far fetched? How does it help?
Discussed in the article. It sharply constrains any hypothesis of communication between the entangled particle generator and the decision to measure a given property. It doesn't change anyone's mind, probably, but it does nail down one of the few remaining corners.
The point was: the checks on this feature of quantum world - that it generates violations of Bell's Inequality - have been numerous, varied, and completely consistent. They have become routine - by now if an investigator failed to discover a violation their equipment or technique or experimental design would be doubted first, long before their results were accepted.