I have heard that SEDE is updated only once a week many times. And for some reason I have ignored this in this instance.
I have still doubts about the last sentence in your answer. At least if I understood it correctly. (As explained in the comment and also here in chat.)
I should find a question which was closed as a duplicate, but the link does not appear in comments/answers/questions neither in the duplicate question nor in the duplicate target to test this.
I.e. the situation where the only link between the two questions is the duplicate banner.
No, it means that if it gets closed as a duplicate it gets an record in postlinks as linktype=3, no need to be mentioned in a post or comment. The example you provided in chat was to recent to have that linktype=3 in SEDE as SEDE us only update once a week, on sunday morning 03:00 UTC — rene12 mins ago
So just to see whether I understand it correctly.
In this particular instance there is one link with PostTypeId=1, because of the comment which was auto-generated when somebody voted to close the question.
This comment was removed in the meantime.
When data in SEDE are updated there will be link with PostTYpeId=3, since it is closed as duplicate.
I assume that the link with PostTypeId=1 will be removed. (Since the comment no longer exists.)
I understood this sentence: Only the direct links are shown in the list. like this: If there is only a link with PostTypeId=3 between two posts and no link with PostTypeId=1, then the questions will not be shown in the list of questions generated by http://meta.stackexchange.com/questions/linked/id.
But I am surprised that I encountered the fact that updates are done only once a week and I did not think that this could explain the apparent dicrepancy.
@rene So in this particular case after the update there will be two rows with 3 and 1. Even if the only place linking between the two posts is a duplicate banner.
Yes, because due to the duplicate vote the generated comment will immediately generate a type 1, the type 3 gets recorded if the question gets closed with a majority vote for the dupe
@MartinSleziak OK, I'll try to rephrase that so it gets more clear and that might also trigger some one who actually has access to the code can verify if my reasoning matches the implementation.