Why aren't duplicate questions automatically merged? Or at least linked both directions?

Question

When a question is closed as a duplicate, it automatically includes a link to the question that it was a duplicate of. In cases where it's closed before it has been answered, this approach makes sense, but what about cases where it generates one or more useful answers before being closed?

In such a case, one (or both) of these two approaches might work better:

1. Automatically add a link from the old version pointing to the new one: this question has been asked again. In some cases the answers to the newer versions of the question contain more up-to-date examples, and if I'm searching for a question and come across the old version, it would be nice to see these new answers as well. This first solution would be quick to implement and maintain and would make it much easier to find all the answers to a question in one place.

2. If the two questions truly are duplicates, merge them. That way any future visitor looking for the answer to the question will find all the answers in one place. This second solution would require more ironing out of logistics, such as

   • once a question is closed, how long do you wait before merging it? (I think it would be best to wait a week or two.)
   • do you use the old phrasing of the question? or the new? or some combination? If a combination, how is that combination agreed upon? (I think it would be best to use the old phrasing and delete the new question. That way, the original poster maintains ownership of his/her question, and later posters are prevented from "stealing" points from earlier posters of the same question.)

To solve the issue of duplicates occurring in multiple locations, a question closed as a duplicate should first gain a link from the old question so that people visiting the old question know to look at the new one for additional answers. The asker of the new question should then have a period of time to edit the question and get it re-opened. If at the end of the period of time, the question hasn't been re-opened, it should be automatically deleted and it's answers should be moved to the duplicate question.

Does this sound like a reasonable idea?

Answer 1

The trick with duplicates is to catch them early so that few or no answers get left against the newer one and can be directed to the canonical question instead.

Questions are duplicates for any number of reasons:

• The asker didn't bother using Search to check if it'd been asked already

Solution:
In such cases it is likely the question is a direct duplicate and would be noticed and closed off quickly before answers get left. If answers are left and they haven't been left already on the original then we can merge, but if they're just repeating the same thing then there is no sense merging posts as you'd end up duplicating answers in the canonical post. Just closing as a duplicate is sufficient in these cases

• The asker did search but didn't find the question

Solution:
In these cases it is likely the reason they didn't find the original is because the wording / language they used when searching didn't match the wording / language used in the original. These ones are ideal questions for marking as duplicates but not merging - it means there are now two different search strings that can be used to find the canonical question, so we'd want to keep the duplicate around to detect this search traffic.

• The user did find the original but didn't like the answers

Solution:
Really in such circumstances they should 'bump' the original question by adding a bounty to it to attract more attention to it. However in some circumstances, for instance if the question is really old and not quite valid anymore then it could be re-asked but brought up to date, while referencing the original question and explaining why it's not useful anymore.

• Two questions are asked independantly of eachother and gain views and answers, and both have accepted answers. Later someone searches the site and sees that these two are infact the same question just asked differently and flags as a duplicate.

Solution:
This is a tricky one. Both were asked in good faith, both have upvotes and both have accepted answers. To merge these questions would result in both the question asker and one of the accepted answerers having their associated rep reduced, which isn't really that fair. If it is a duplicate then it's better to mark as a duplicate and keep open, possibly manually adding a comment to the canonical one linking back to this newer one.

As for your suggestion about 'backlinking' from the canonical to the newer ones, that is a sound idea in principal, but it can easily become quite a headache in practice when a question has been linked to numerous times.

For example, the 'Must Read Interface Book' question gets linked to quite a bit. A quick Google search for site:http://ux.stackexchange.com "duplicate: Must-read User Interface Book?" returns 22 results (and I'm not even sure that's the best way of identifying how many incoming links a question has from its duplicates). Having to provide a list of 22 linked questions in such a manner would be quite a PITA, and it's not likely people are going to read through all of those posts.

So yes, in spirit I would agree with you that it'd be useful to know if a canonical question is a canonical question for other posts out there but it's a bit tricky to figure out how to convey this.