The general rule is that a new question is a duplicate of an existing one if an answer to the existing question would also answer the new question — even if no such answer has yet been given, if it’s obvious both questions are asking substantively the same thing such that an answer to one is an answer to both.
Marking a question a duplicate is essentially saying, “Somebody already asked this. If that other question doesn't solve your problem, please clarify your question to explain how it's different.”
There are cases where the numbers don’t matter. We see various reformulations of the water jugs problem that differ in the sizes of the containers and the starting/goal amounts of water, for example. But a general solution exists for these puzzles and has been posted as an answer, so basically any of the standard formulations of such puzzles can be closed as a duplicate of that one.
Other puzzles, such as the more or less yearly “form the values 1 to 100 from the digits of the new year, using only $(set of operations)” ones, don’t have a general solution. The rules tend to differ slightly (digits in order or not? initial concatenation allowed or not? concatenation allowed as an operator? which operators are allowed?) and the solutions depend on the usable digits. A solution for 2012 may work in 2021, but we weren’t around then :) so it’ll be a long while yet (2051!) before we land on a year whose digits are a reshuffling of a year that’s been used in a formation-of-numbers puzzle so far, or (2023->2032) is likely to be in the future.
Since these questions thus are not answered by another question, there’s no reason to mark them as duplicate.
Having said that, genre fatigue happens when too many of the same type of formulaic-looking puzzles are posted in too short a time. They are, usually correctly, viewed as being low-effort rehashes of the same type of puzzle with no effort required by the poster beyond exchanging one value for another. Downvotes are the inevitable result, and probably the correct one. So, as @GentlePurpleRain noted in a comment, these indeed tend to be self-correcting problems.