We've gotten used to the situation here where someone has an idea on how to improve things and we all race to be the first to say, "We can't change the software."

This is not always the correct response. Trackbacks to the arXiv were implemented with no changes to the underlying software, and many small features (eg citation links) can be implemented in javascript (another eg: all MathJax support).

One thing I should have mentioned somewhere earlier is that the SE has agreed to let us continue to control a div in the footer, so we will not lose the ability to implement javascriptable features.

@quid: You're probably right that if it takes a while for it to become clear that the thread should be on discuss instead of meta, it's probably best to start a fresh thread. But especially early on in the two-meta experiment, I think making it easy to move between the two will make it easier to establish what belongs where. Even if it's best to start a fresh thread, I would still rather eventually delete (or close and lock) the thread on meta and move the data to discuss. Otherwise it seems like it would be confusing to new users who are trying to figure out what the three sites (main, discuss, meta) are for.

So, not to (be able to) migrate seems like an advantage to me.

Regarding the identity problem, one possible solution is following: MO moderators can confirm the association of a discuss.MO account with the corresponding SE account by checking that the OpenIDs do match. The OpenIDs are not public, but community moderators can see them.

• The SE-style meta is terrible, so with two sites, meta.MO (the SE-style meta) will be a ghost town and all the action will be on discuss.MO.
• The two sites will be useful for different things. We have a strong core community of people who care about meta; they'll figure out how to best use the two sites and newbies will follow suit.
• Discussions here are petty and counterproductive, so with two sites, the SE-style meta is where the real action will happen and discuss.MO will either be a ghost town or a place for angry people to rant.

To put it another way, supposing you believe one meta format is superior (at least for certain things), will the existence of the other meta format be a long term serious impediment to the correct function of your preferred meta? For me, the greatest concern is that the SE-style meta will destroy the culture we have have here. But if we can eliminate the technical barrier between the two, I don't think this will be a problem.

@Scott and Andrew (and anybody else, I guess): We should solve the migration and identity problems now, if only to get a clear idea of what is involved. Really, it would have been a good idea to solve the identity problem a long time ago.

For identity, is it enough to add a column to the meta user table (I assume such a thing exists somewhere) for MO user id number and then check that the md5 of the email matches the gravatar hash of the MO user? With this system, do we have to worry about somebody guessing the email address of a given user, thereby stealing their meta identity? What would need to happen to be able to use the login cookie from MO? That would clearly be preferable. At the very least, we would need the SE team to change the domain of the cookie from "mathoverflow.net" to ".mathoverflow.net" (which they may be reluctant to do), but even once we can actually access the cookie, it's not clear what to do with it.

But the right home for the first three is almost always meta.SE, not meta.whatever.site.it.happens.to.be.

For what it's worth, an announcement was sent out about this. The SE team checks site metas often enough that it's not necessary to post to meta.SE for network-wide issues.

Each question on meta.SE is required by software to have one of the following tags:

bug, feature-request, support, discussion.

The meta.SE is good for the first three, which are not discussions.

But the right home for the first three is almost always meta.SE, not meta.whatever.site.it.happens.to.be. We've gotten used to the situation here where someone has an idea on how to improve things and we all race to be the first to say, "We can't change the software.". Exactly the same thing happens on the SE sites! Someone has a good idea, we say "We can't change the software.". The only thing extra is that we say, "You have to take that to meta.SE." and my experience of that is that ... nothing happens. Usually, all the Good Ideas have been had and roundly trounced years ago so all new ones are marked as duplicates. It isn't quite 100%, there is a little space for these posts, but it's close. For example, sometimes it's hard to find information on meta.SE (make that "always"!) so it's easier to ask first on the per-site meta to see if anyone else knows.

The other thing that the various posts have brought to my mind is the voting. I cannot conceive of a situation where voting on the per-site meta is a must-have. It isn't a good way of getting a poll of the users of the site: only a very small minority visit meta, and they will tend to be of a particular type so not in any way representative. Kaveh's point about votes is important: they tend to polarise discussions.

Lastly, even when we've been discussing desirable and undesirably behaviour on the site, it's been rare that we've gone so far as to say, "Let's lay down the law on what must be done". A vote would give some proposal some legitimacy that I don't think it should have. When we've argued about what questions should and shouldn't be allowed, the discussion has almost always been to try to persuade others of ones point of view, never to say "Everyone must close questions according to this rule". When it's clear that I'm in the minority on one of these discussions, I never feel as though people are saying that I can't behave in a certain way, just that when I do so I'm likely to find that I'm acting counter to most other people.

(To be clear, I'm talking about things like deciding which questions to close, not clear inappropriate behaviour.)

Okay, it wasn't lastly. Lastly, on a technical note, it could be that we have a space in the user account here for a link to a person's SE id. For verification, one could use the gravatar hash, I guess. Ultimately, moderators of both could see the full profile and check email addresses. It would also be possible to make this place accept OpenIDs for login purposes. If SE were willing to allow this site to read its login cookie (if it were possible, I don't know) then that could be used for authentication.

The SE API will make it easy to gather and parse a thread for purposes of moving it, but there are some other hurdles:

• the thread should be closed (and probably locked) to keep new posts in one place,
• the data need to be imported into discuss (is this easy?)
• for smooth migration of posts, users must have persistent identities

The last one is the one I'm most worried about. OpenID strings are not included in public database dumps, and even if they were, Google OpenIDs are crazy. However, gravatar hashes are public, so we may be able to use them to at least make a persistent identity for anyone who submits an email address.

So far I've been very hesitant about having two metas, but being reminded of the bug/feature-request/support/discussion dichotomy imposed on SE metas, I'm becoming more relaxed. It seems that keeping the first 3 on a SE meta, while encouraging (forcing?) discussion to take place on a forum, might actually be workable.

Discussions happens on meta.SE in comments sometimes, but that is not good because of the restrictions that apply to them, they can not be more than 600 letters, so sometimes people post another answer explicitly referencing the other answer and stating that their answer didn't fit in a comment (SE people do this quite often).

I think having two meta is not a bad idea. We essentially implemented something like it on cstheory as meta policy. We have a post about an issue where discussion happens and votes are interpreted as "like" and "dislike". Then when we want to decide about a policy we use another question (e.g. this) where votes are interpreted as votes on the policy.

Each question on meta.SE is required by software to have one of the following tags:

bug, feature-request, support, discussion.

The meta.SE is good for the first three, which are not discussions. Most meta-questions with discussion tag do not need voting. It seems to me that in an off-line setting we don't ask people to vote in a discussion, maybe in a debate or making a policy decision, but not in a discussion.

I think a forum format like meta.MO is better suited for a discussion, the meta.SE format is better for questions about the site (support tag), bug reports (bug tag), feature requests (feature-request tag). meta.SE format is also good for taking votes to make a decision (but I think it should be used carefully so it does not cause friction of the kind that would cause some to be less active on meta). Discussion questions on meta.SE can be posted on discuss.MO (unless they are polls about a policy), the other three kind of questions on meta.MO.

Migrating from meta.SE format to a meta.MO format should not be difficult: ignore the votes, sort the comments and answers by time-stamp, add @[user-time-stamp] to the beginning of each of comment/answer from meta.SE.

(We can even write a user-script that would show them in a threaded fashion, there is already a user-script for displaying comments in that form).

In summery, I think having meta.MO and discuss.MO is a good idea, and unlike others I don't think it would cause confusion and is practically implementable.

ps: by meta.SE I mean meta of a site, not the general meta.SO.

While the SE engine is absolutely horrible for extended discussions (and the chatroom, while great for discussions, is bad for documenting threads and such), there are certain things the SE engine excels at:

• The ability to easily mark an item as resolved (by accepting an answer)
• Simple questions that does not require extended discussions ("how to use the website"; "something's wrong with my account"; "Wasn't there a question about blah? Where did it go?") are served very well by the SE platform.
• Questions phrased as a poll between several options. (So we can use the votes to tally opinion.)

And I see this split somewhat differently from Deane. I don't think this is a split based on "content"; but more implementing two facets of Meta using tools more appropriate for the tasks.

On point four though, I'd just like to point out again that we currently don't have much in the way of ability to modify the product to suit our specific needs.

1) Having a meta.X.SE run on the same platform as X.SE is an a priori bizarre decision. The whole philosophy of the SE engine is that it is tailored -- perhaps even socially engineered -- to be optimally successful at a very specific kind of internet interaction. I agree with this philosophy. But then it gets turned on its head when the same platform is used for the very different kind of interaction that occurs on the meta sites. We get a community discussion where everything is sorted by votes and the top answer is the one selected by the questioner: huh?? In some cases it is not even clear what upvoting/downvoting means. On the main site, upvoting a question or an answer is supposed to be an assertion of quality. On the meta site, the most standard meaning seems to be "upvote if you agree, downvote if you disagree". But this reduces discussion to a bunch of people holding up thumbs up or thumbs down. Many times the questions themselves are downvoted, but what exactly does this signify? How do you disagree with a question??

2) I think much of the point of any meta site is to function as an outlet for negative/petty/emotional/off-topic interactions that one doesn't want to occur on the main site. This particular meta works very well for that, I think. (Of course this means that many threads on meta are not going to be very pleasant to read: they're going to be a bunch of people sniping at each other, or one person saying something and then a lot of other people saying "How could you say that? That's actually offensive.") Certainly negative / petty / emotional interactions occur on meta.X.SE sites as well: they just come threaded differently. One could argue that the ability to vote (at least, upvote) on every single thing anyone writes can exacerbate this confrontational aspect. If someone expresses their opinion and comes back to find their answer marked -7, then a lot of people have simply piled on their disapproval without advancing the discussion in any way.

3) In order to function properly as an outlet, the meta site needs to be one step away from an absolute free-for-all, which I think is how this site currently operates: anyone can say pretty much anything they want once or twice or thrice. When it becomes abundantly clear that the discussion has spiraled completely out of control and away from any actual discourse, then the moderators may close the thread. On meta.X.SE questions, answers and comments are too routinely removed by higher-level moderators (i.e., members of the SE team, not the local community). When this happened to me it left a bitter taste in my mouth that I have not completely been able to wash away.

4) Having two different meta sites for MO seems rather complicated and confusing. It also doesn't strike me as much of a concession on the part of the SE team. It seems to evade the main question we've been discussing concerning the migration: to what extent are they willing to modify their product and practices to suit our specific needs? What's the answer to that question shaping up to be, by the way?

As for finding out what people think, and the voting issue, as I've said: it would be easy to add some sort of voting to this forum. But I don't see the need for most things. Why would I need to know how many people agree with JDH? It is enough that JDH has that opinion and that it is a reasonable opinion. Knowing that 10 people have clicked on a button wouldn't add anything to the notice I would take of it. Particularly as I wouldn't know if those 10 people were 10 people who's contributions I respected or 10 random stop-bys from SO.

To counter JDH's data point (slightly tongue-in-cheek) one of the reasons that I don't participate in maths-SX is the heat and arguments that rage through its meta that do not have a cooler head countering them.

Back to the topic of the thread, if someone who is supporting the SE form of meta can tell us what it will cost to use it (in terms of what we would give up by abandoning this form of meta), I think the discussion can take a more focussed turn.

Gerhard "Ask Me About System Design" Paseman, 2011.09.16

Meta.MO is where we remember that the other people involved are actual people. We may not agree with them, we may never agree with them, but it's where we find out what they think so that we can act accordingly. The potential for misunderstanding and upsetting people on MO is large and meta.MO acts as a place to mitigate those aspects. On SE/SO I don't have the same empathy with the other users that I have with the participants of MO and that's because there's no place to interact at this level.

When MathOverflow started, Anton and I had many discussions on meta as to what was and wasn't appropriate here. Those discussions would not have been possible on SE/SO! But without those discussions, I wouldn't have changed my mind and agreed (eventually) with Anton.

The SE/SO model imposes a "top down" mentality on the site. Decisions are taken "on high" and meted out to the lower orders. MathOverflow does not work like that. It is much more a community-led site than the other SE sites are. I feel that I have a reasonable chance of getting Anton or Scott to hear my point of view. I have no such feeling with Jeff, Joel, or any of the others on the SO team. And that's not entirely due to personalities: whenever I have interacted with one of the team then I've found them to be largely pleasant and reasonable. The difficulty is getting the chance to interact with them in the right manner.

So for me, meta.MO is what frames MathOverflow and if we change it for a different system, we'll significantly change the character of MathOverflow.

A couple of other (minor) points:

1. Yes, there are some things that the SE/SO model does better than the forum model. But I think that splitting is a bad option because the split would not be clean. Many of our most useful discussions started out as something different.

   It would be possible to modify the forum set up to take into account these things. This particular forum is a very basic installation, there are loads of features that can be enabled to change how things behave. The forum does some things fantastically, and a few things badly; but the SE/SO model does a few things okay, and most things badly. My conclusion from that should be obvious!

2. The SE/SO "chat" feature is irrelevant for this. The TeX-SX chat rooms are fun and useful, but they are for real-time interaction only. It is very hard to read a transcript as there is so much noise, and once something is far back enough that you need to be reading a transcript (rather than the "live" data) then everyone's moved on and forgotten what the original topic was. If someone happens to be there then it can be very useful, but otherwise it's not anything like a forum.

3. There's a famous diagram somewhere showing Jeff or Joel's view of SO as lying at the midpoint between a wiki, a forum, and a blog (or something like that). The argument is that it has the best features of all of them (I disagree!). It's a little ironic, therefore, that all the SE sites are now getting community blogs! Why the opposition, therefore, to a community forum?

If it were up to me, it would be "keep meta or we don't migrate"

The SE/SO version of meta is a complete and utter shambles. I can understand why they have it, and for a certain model of their software I can see why it works, but I think that that model does not fit MathOverflow and we would find it to be actually harmful to the site.

Here's the longer version:

First, my experience. I'm not putting this to browbeat anyone into submission but to explain where my views are developed from.

1. I'm the system administrator of the nLab/nForum set up. I became that because after then nLab was set up, there was a need for discussions and these were taking place in the comments of the nCafe. I thought that this was not a good system so I proposed and set up the nForum. The nForum acts as a "meta.nLab".

2. I was one of the earliest users on MathOverflow.

3. The existence and format of meta.MO was my suggestion, based on my experience on the nLab/nForum.

4. I've been involved in various StackExchange sites, most notably TeX-SX. I was a pro-tem moderator there and have been very active throughout its life, both in the day-to-day activities of asking and answering and in the more "meta" activities. I regularly use the main site, the meta site, and the "chat" sites.

It is my opinion that TeX-SX works well despite its meta and not because of it. The SE/SO model of meta is very good for getting a snapshot of people's opinions on a particular topic. I've used it for such from time to time. But it is very bad for actual discussions where the expectation is that you will try to persuade someone of something. Whenever I've tried to use meta.TeX for that, I've found it frustrating.

What do I not like about the SE/SO model:

1. The assumption that everything important starts with a question. And not just any question, but a question that can be given an accepted answer.
2. The assumption that every other contribution has to be an answer or a comment on an answer. Contributions are contributions. Some are in reply to others, some are not, but it is not reasonable to force it in to an either/or.
3. The lack of true linear ordering. Even ordering the answers by time doesn't help as the comments are not properly interleaved. Also, one has to decide "is my contribution a comment on someone's answer or a new answer?". It is important to be able to know what someone had read when they posted some contribution.
4. The presumption that there is a correct "answer". Even when we reach a consensus on meta.MO, there is a huge difference between that and marking something as an accepted answer! And often, we "agree to disagree".
5. The presumption that votes are important. Actually, I can see a place for a sort of voting system. We do get a few "I agree with X" comments here on meta.MO. But these still add something in that we can see who said them. So I would have voting if it could be not anonymous (by which I mean you could see which account was used to vote). Also, I would make it not possible to sort by votes.
6. The ability to edit others contributions. On the meta site, this is almost never acceptable. Certainly not just by others who have been around long enough to have earned enough reputation.

That is, discussion happens on their metas, but it seems like it happens largely in comment threads, which is awkward (@active SE 2.0 users: please comment on this!).

This agrees with my experience. Ideally people take stances by giving an answer first, but sometimes all discussion takes place in comments to the original question instead.

For the solution to work, I think it would be important to have clean ways to migrate stuff from meta.MO to discuss.MO (by the time it's clear we should migrate a thread, it may have grown cumbersome), and I think we would have to put some work into making it easier for people to post here (current meta.MO, future discuss.MO).

Related to this, a good thing about site metas is that you can migrate from the main site to meta easily.

What to me is the key difference (and essentially the online one I can really see) between SE-style meta and MO-style meta is voting. [I should say I never used SE sites, but in recent times looked around quite a bit over there.] I think this changes the spirit quite a bit.

If one could turn off voting (at least selectively) I think the two are essentially equivalent.

Sorting by 'oldest' is supported (and also sorting by newest/active can have some use, and in the absence of votes there is no other way to sort), and so the sequence of 'answers' in SE style would be just the posts here, and the comments would allow one level of threaded-discussion which I'd actually consider as an advantage for short comments on somebody elses point, say to express punctual agreement or disagreement.

So spontaneously, SE-meta without votes or with the option of having them turned on or not (default would be something to discuss then) seems fine to me.

I just saw Thierry's comment: thinking about it, the visual similarity could be a problem, I agree. Would it be possible to have a different color-scheme or some other layout-changes to get some visual-gap?

ADDED: Slightly off-topic question but in view of Steve Huntsman's comment that I like. Could somebody in the know comment on this chat-feature of SE. Is this used on math.SE? If so what is happening there?

I would like to add that (in my very limited experience) the comments on MathOverflow and math.stackexchange push the restrictions toward a discussion format. If I were allowed only to post answers, or very brief comments, I would not contribute to the site nearly as much as I do now. If future versions of the MathOverflow platform are considered (and I am thinking/writing on this matter), I propose that a discussion type element be integrated so that those who wish to contribute/view such discussions associated with a question be given an appropriate filter, and those who prefer the traditional, briefer style be allowed that form of display.

Gerhard "Ask Me About System Design" Paseman, 2011.09.15

Also, perhaps we should move this discussion to meta.meta.mathoverflow.net.

• SE 2.0 is built with the assumption that every site has an associated meta site running the SE engine. They are adamant about maintaining this level of uniformity (and I think this makes sense).
• We like our meta. Our community standards have been built largely on old-style discussions which involve many back and forth cycles between participants. A Q&A format is made not to easily allow such back and forth to occur. For example, this thread (which is a solicitation for opinions and responses to those opinions) would be a poor fit for the SE engine.

A proposal which is on the table is to move this meta to discuss.mathoverflow.net and to have an SE-style meta at meta.mathoverflow.net. I think this is a reasonable proposal, but I don't claim to know what the optimal meta setup is. Here is a summary of my thoughts on the issue.

It is certainly true that SE 2.0 sites manage to reach some consensus about their community standards. That is, discussion happens on their metas, but it seems like it happens largely in comment threads, which is awkward (@active SE 2.0 users: please comment on this!). I think they don't know what they're missing. On the other hand, there are also things that we are missing which are provided by the SE-style meta. The SE-style meta is certainly better for things like UI questions, bug reports, feature requests, and extended FAQ. Moreover, SE-style metas are much better integrated with their main sites. Here, it is not uncommon to see posts that start "I finally decided to register so I could post here".

If we have discuss.MO and meta.MO, we will in some sense have the best of both worlds. I suspect that our usage habits would eventually dictate how the twin metas are to be used. However, there is some overhead cost to having two meta sites. For the solution to work, I think it would be important to have clean ways to migrate stuff from meta.MO to discuss.MO (by the time it's clear we should migrate a thread, it may have grown cumbersome), and I think we would have to put some work into making it easier for people to post here (current meta.MO, future discuss.MO).

Another thing I'd like to do is entertain the possibility of pushing for the SE-style meta to allow the sorts of discussions we have here. The SE team is not very excited about this possibility, since they've had great success by limiting discussions. However, I think minor changes to the engine may produce a satisfactory solution. For example, suppose it were possible for a thread to default to (reverse?) chronological sort order if the top post has the [discussion] tag. I should emphasize that this possibility is not currently on the table; I'm only entertaining it personally. If this feature existed, in what way would the SE-style meta be lacking? Would we still want to keep this meta around?