Before closes as "unclear what you're asking", I'm asking the SE overlords and/or mods, is this is a fair summary?
Some comments on questions and one answer on the Meta discussion Can we be more careful in marking duplicates on questions that are related but ask for or about different things? seems to suggest certain users here are labouring under the mistaken belief that if two questions are different, but an older question's answer contains content that partially answers the newer question, that's grounds for closing as a duplicate.
It isn't. A duplicate means:
This question has been asked before and already has an answer
When is something a duplicate?
This has been discussed extensively on the main meta and official blog. Details below, but to summarise:
- Closing as duplicate is for when the question is the same. The answers should match, too - but checking the answer matches is described by SE staff as a secondary "litmus test" after identifying that the question is the same:
- Is the question the same? If no, don't dupe, see below for alternatives.
- If yes, does the existing answer also answer this question? If no, reconsider question 1.
- If yes to both, dupe.
- Canonical questions to general types of common problem are encouraged when there's one pattern that answers many slight variations on a question. Encyclopedic "massive canonical Topic FAQ" "tomes" covering entire topic areas are discouraged because they are too broad for SE, are bad for SEO, aren't user friendly, and because abuse "chokes out new information"
It really is about the questions, but checking answers is still important
SE staffer shog9 recently posted an answer for another SE site where people tended to dupe different questions that share an answer but were different questions:
...I would strongly recommend using answers as more of a litmus test than as a policy: if you're already pretty sure the questions are duplicates, testing the answers of one against the other can easily confirm your suspicions. But don't close completely irrelevant questions as duplicates of one another simply because there's an animated gif that happens to apply to both...
The simple solution when answers would be similar
If an older question's answer contains content that answers or partially answers a newer question that is not the same question or problem, the best approach is usually to post a short cross-linking answer like this:
You have a similar problem to people who do X. The advice on our question for them is:
[quote from conclusion]
This is based on [short quote or summary of gist]. Since you do Y, not X, you are different in that [...], but this advice still applies, because [reasons].
This way, people aren't left scratching their heads struggling to find the relevant bit when following the link, and if someone else knows more about this specific question than you (heaven forbid!), they're not prevented from posting a better answer that shares relevant detail that wouldn't be relevant to the (non)-dupe.
Sometimes, questions can be broadened so one canonical pattern can solve a range of questions...
From an older post from SE staff member shog9.
When you see a question that seems like it might reflect a common problem, don’t just answer it to get a few points. That doesn’t make the Internet any better. Instead, help us build up a library of canonical questions and answers that are more generic versions of the same question, and then start closing all the exact duplicates.
...
All that said, be careful not to go too crazy with this idea. The example you gave was a very poorly-asked question with a very simple (and common) problem - it's easy enough to find another question that covers the same problem. But some problems are not so simple, and it can be harder to find an existing question that covers the exact same issue.
In other words, when duping to a canonical general question, it still must address this exact issue, not simply be related background info. If it's relevant but not a match, do something else, like post a cross-linking answer with a short explanation (see below).
They must be based exactly on one common underlying problem or pattern
Shog9's first of several related reading links expands on this a little with a quote from an official blog post. The examples are very programming specific but it's not hard to see how they relate:
If you’re going to close a user’s question as a duplicate, it has to be a real duplicate. For example, if a user asks, “What does the IP address 128.0.1.1/24 mean?” it’s OK to close that as a duplicate of a more general question like “What do IP addresses of the form a.b.c.d/e mean?” But it’s not OK to close it as a duplicate of a twenty-seven page guide to netmasks. That’s the moral equivalent of saying “RTFM.”
Stack Overflow is not meant to be a library of reference manuals. It’s supposed to contain the same information as a library of reference manuals, in the form of millions of questions and answers. Combined with Google, that gives us the magical power of a library of reference manuals you never have to read! It’s like, you got to the library, and there’s a wizard there at the door, and you ask your question, and, instead of being told to read a book, you just got (are you sitting down?) the actual answer!
They're not a tool to arbitrarily reduce the number of questions, throwing out nuance and depth: having "millions of questions and answers" is a good thing, so long as they reflect different problems (even subtly different problems). They're for cases where the underlying problem is exactly the same and is solved completely by one pattern.
They should be used with care and shouldn't get out of hand
That June 2016 Shog9 update to the policy over on the SciFi site (one of the few sites to also have aggressive dupe-closing of related questions) clarifies the drawbacks and limits of canonical questions:
The idea of a massive canonical Topic FAQ is persistently alluring; instead of having 50 questions about a "sorting hat", you could just have one question with a few dozen huge answers! However, this quickly becomes impractical:
- Massively broad questions don't rank as well in search results as specific, focused questions.
- Folks with specific, focused questions tend to not read massively broad FAQs even if they do find them.
- Finding specific information among multiple answers to massively broad FAQs is troublesome.
- Remembering which information is even contained in these tomes is difficult; eventually, folks just start to assume that they contain everything and close new questions without worrying whether they're actually answered or not... This chokes out new information.
There have been many, many, many cases highlighted on meta and in comments recently where aggressive dupe-closing "chokes out new information" relevant to a particular case, resulting in lower quality content.
It's right and proper to allow many variations between similar questions when those variations reflect real variation in the underlying real life problems reflected by the questions. This allows answers to have more nuance and higher quality. As a 2010 official blog post put it:
It's far more common to have many subtle variations of a question. I think that's OK, because that's how the world works. Trying to shoehorn a bunch of semi-related things into one arbitrary container in service of some Highlander-ish "there can be only one" rule is ultimately harmful. Remember: while there are aspects of wiki to our system, we are not Wikipedia. There is not one canonical question about every possible subject.
What if a new question isn't quite covered by a closely related existing canonical?
Bloating the canonical to fit a case that it's pattern or technique doesn't currently fit would clearly risk heading towards the massive canonical Topic FAQ tomes that are warned against. There's nothing awkward or wrong about having edge cases that are like a common canonical pattern, but with circumstances different enough that this doesn't solve them. Even the SE network's #1 ranked user Jon Skeet can have a question that is like a common canonical problem type but subtly, importantly different.
Let's use a recent example, where we have a great canonical resource on travelling with dual citizenship, and someone has a delicate problem where it'll be very difficult for them to do what the canonical answer suggests, where duping to the canonical question is like telling someone to read a (short, well-written) manual that doesn't actually even address the details of their specific problem.
shog9 gives two suggestions, and the second fits:
- Post an answer that relates a different question to this one. I'm doing that right now (though with a blog post instead of a meta question as my source). Sometimes, it's hard for an asker to understand a solution presented elsewhere without something that relates the problem being solved there to the one they're having - a short answer that introduces, summarizes and then links to an answer elsewhere can provide the best of both worlds: a specific, focused answer to a semi-novel question and the refined knowledge of other members of the site, ranked and reviewed over the years.
...except of course in this specific example, the canonical doesn't contain the answer to their question, it only gives the background to why what they're trying to do is problematic, but you get the idea. Cross link, quote, explain why it relates, add missing details.
Example dummy answer
First, you should know that you shouldn't be travelling this way. [Here's our standard advice for similar cases](link to canonical).
Obviously it's too late to do this, since you've already booked tickets a different way, so you need to decide which is more important between rebooking your flights, or risking [x/y/z].
[add any further knowledge relevant to this case].
Even if you're only adding a little case-specific nuance beyond what the canonical says, this is better than closing, because someone could come along later who does has specific experience or professional knowledge of the asker's exact situation ("The authorities do X checks, and people get caught doing what you want to do all the time", or "You can safely do this because of A B C specific details, but only if you do X, Y and Z"), and they are not barred from offering a better answer. New information is not choked out.
To summarise:
- duplicates should be the same question (and also check that answers match up as a "litmus test")
- canonical questions and answers should be outlines of patterns that fit many cases (equivalent to “What do IP addresses of the form a.b.c.d/e mean?”), not user unfriendly, SEO unfriendly "massive canonical Topic FAQ" "tomes" stuffed full of every variant on a topic or theme.