Mass Creation and the Liturgy of the Profane
neither mandatory nor forbidden
Of the many things people who don’t edit Wikipedia think they know about it, the biggest is article deletion. Much of the project’s internal workings are hidden, but deletion discussions are marked by huge red banners atop an article, and plenty of people jump in when they see them. Plenty more people know firsthand the experience of having tried to contribute to the Sum of All Human Knowledge and having their suggestions turned back on them. These people tend to come away with the idea that Wikipedia is rabidly deletionist. They’re not entirely wrong, but they’re missing half the story.
Wikipedia has a gauntlet of inclusion standards, but the one that matters most times and places is the General Notability Guideline. This declares that subjects are “presumed to be suitable for a stand-alone article or list when it has received significant coverage in reliable sources that are independent of the subject” (emphasis original). This is not an arbitrary group (well, mostly — the last two clauses go down weird rabbit holes). Wikipedia articles are written entirely from sources, without the opportunity to add one’s own discoveries or reinventions; if something lacks significant coverage in reliable (a term-of-art with complex relations to its real-world use) sources, the article will be very sad. People disagree quite substantially on what sources are good enough to demonstrate this state, though, which is what gets you the real controversy.
Alongside the GNG, there are Specific Notability Guidelines for a diverse array of subjects. SNGs try to be vaguely accordant with the GNG, in terms of producing broadly similar “allowing as many non-sad articles as possible” outcomes. Many SNGs are as written more lax — for instance, the notability guideline for books is quite forgiving about how in-depth a review needs to be. Others are more strict; many have complained about how standards are applied to fictional characters and phenomena, and I personally think the notability guideline for songs is a bizarre fit of bureaucratic deletionism. Mysteriously, the project is often more active in areas where notability guidelines are generous. Must be one of those coincidences.
The relationship between SNGs and the GNG is complex and contradictory. A significant minority of editors believe SNGs should be flatly abolished, that one-size-fits-all applies to inclusion standards. Many SNGs are nominally subservient to the GNG as “presumptions of notability” — but you’ve just noticed the GNG is also a “presumption”, isn’t it? Some, such as the guideline for academics, are explicitly orthogonal to GNG; that guideline is a lightning rod for disputes, being either way too high, way too low, or both depending who you ask.
Non-editors who are exposed to a fraction of the eldritch god’s face tend to ask why we need all this. Why not be generous? It’s obvious wherever we can see it that people are encouraged when their edits are kept, and discouraged when they’re reverted. Why do we have any of these guidelines in the first place, and why on Earth would they keep getting stricter, as evidence mounts that strict guidelines discourage new editors we desperately need to recruit?
At last count, Wikipedia has around 6.7 million articles. Anyone who was a bored schoolkid in the internet age remembers hitting Special:Random a ton of times, escaping class and school internet filters by riding the Information Superhighway, probably learning more in the process than you did in your compulsory miseducation. Cast your mind back and remember how many of those articles were stubs.
“What’s a stub?” (I’m reminded of a friend who I infodump all my wikipursuits to seeing an editor use the terminology “fleshy stubs” and finding it rather unpleasant.) Wikipedia has a beautifully arcane internal article rating system that was made for — get this — triage of what should be included on a CD-ROM version. I love it, because I have exactly the kind of problem where you obsess over rating schemes. Is it relevant in any way to a version of Wikipedia that no one is ever going to try put on a CD-ROM? Who cares?
One reason the assessment scale has long outlived its origin is that it’s an intuitive way to classify articles. Some of the ratings on that table aren’t ‘real’ anymore (A-class is functionally deprecated and only in use by two wikiprojects), but the general scope of the scale is much the same thing you see in a Metascore. Some articles are fantastic, some good, some bad, some atrocious. Many are mixed bags.
Stubs are the very bottom of the scale — in most cases, barely more than a sentence or two. Here’s a handful pulled from Special:Random:
Ricardo Acuña (born 23 January 1971) is a Mexican judoka. He competed in the 1996 Summer Olympics.Vera Yakovlevna Komisova, née Nikitina (Russian: Вера Яковлевна Комисова bzw. Никитина, born 11 June 1953) is a retired hurdler. Her career highlight came in 1980 when she won the Olympic gold medal.Səməd Vurğun is a village and municipality in the Shamkir Rayon of Azerbaijan. It has a population of 1,433.[citation needed] The village was named after Soviet writer Samad Vurgun.Adelpherupa terreus is a species of moth of the family Crambidae. It is found in Madagascar.Lembah Impiana is a suburb of Kulim District, Kedah, Malaysia. the suburb has its own commercial center, it also treated as one-stop center for food, because several cafe and restaurant can found there.Three of these articles have exactly zero sources. None of them are more than three sentences long.
For some subjects, a stub is a wholly legitimate addition to a reference work. If a clearly notable historical event has little in the surviving record, you may as well write up what you know, and photographs of a geographic feature are worth well over a thousand words. For others, it’s less clear that any encyclopedic value is being provided.
If you recall the last time we were here, you might notice two of these stubs are biographies. Knowing everything we do, why would articles about people end up in that state at all, let alone at any meaningful rate?
Many of Wikipedia’s earliest articles were straight rips from public-domain encyclopedias, most notoriously the 1911 version of Britannica, which still lurks mainspace. The first attempt at getting a bot to import these was just months after the birth of the project, when a contributor imported some great number of articles from Easton’s Bible Dictionary (1897). This caused no little contention on the — man, I love the early Internet — the Wikipedia mailing list, where people objected to such results as a bustling town being described as a settlement of a hundred people. Nonetheless, the attempt as a whole saw more praise than criticism, and such articles have mostly expanded into true modern reference works. I recently saw our well-developed article on Anaximenes of Miletus nominated for the Good Article peer review process, and found it was created in 2001 as such an import.
Following this, Rambot was created in late 2002 to automatically import articles about American towns and counties based on census results. “Places where people live” are trivially obvious coverage areas for a reference work; definitionally, people are interested in reading about them, and even entirely database-sourced articles can have substantial and in-depth information given how comprehensive census results are. If you’ve ever noticed that articles for small towns in the US tend to have far more detailed explanations of their 2000 census results than the later censapodes,1 that’s because of Rambot.
Rambot was not wholly uncontroversial. Some people felt it was “cruft” — a term-of-art that I’m told means something other than “I don’t like this”, but no one can agree what. Others pointed to legitimate if mostly minor errors, such as false precision in percentages and bizarre presentations of small towns. Nonetheless, the articles it made were as a whole reasonably well-developed, even if the passage of time has made them less valuable. You learn far, far more about a place from its census results than from the “History” sections editors focus on that no reader in the entire world has ever looked at.
This successful early attempt sparked the question of whether it could be used elsewhere. In 2008, a member of the ClueBot family (Wikipedia’s anti-vandalism bots, which serve to automatically revert twelve-year-olds replacing articles with “U ALL SUCK COCKS”) was repurposed to create a few thousand single-sentence articles on asteroids. This was also…not uncontroversial. The astronomy wikiproject was unhappy to see these articles of minimal informational value, unlikely to be expanded and more likely to pose maintenance burdens, taking up residence in their topic area.
At the time, there was no real notability standard for astronomical objects. The botop worked by extrapolating notability for populated places and geographical features, assuming such objects were notable by the nature of their existence. This works just fine when you’re making articles about astronomical objects anyone might conceivably want to cover. It falters when you’re working from a database dump of Every Asteroid Ever. It turns out that, given a database dump of Every Asteroid Ever, most of them can’t have articles that serve a higher value than making Special:Random worse.
Astronomy editors got to work. They didn’t want a repeat, and they wanted some way to get rid of these articles. If they were all notable, the only option they had was to render them non-notable. After extensive discussion, they worked out a guideline that roughly corresponded to “things people might conceivably want to cover” (e.g. objects visible to the naked eye, objects listed in major catalogues, objects people have gone out of their way to study) and turned all the bot stubs into a list.
Assuming some sane series of events, those articles never would have existed. It’s self-evidently ridiculous to throw a database dump into Wikipedia in the form of a ton of individual tiny articles that can’t possibly not be tiny. More troublingly, it’s possible to consider data integrity consequences. If you aren’t iron-certain that your database is good, and lacks any intentional red herrings/evil bits, you can perpetuate serious errors across the whole world — Wikipedia is a major data source for reusers, and exactly these problems regularly come up with, say, Google Maps results.
Of course, “database” is a broad term. Many databases are rather in-depth, including the aforementioned census records. You can write a full article based on census results in a country (this includes the US and Australia, at least) that collects a lot of decent-quality data about its citizens. But when “database”-written articles develop a reputation for being one-line stubs, people develop the assumption that articles written from “databases” must be one-line stubs. The transitive property involved here is seductive thought, but not accurate thought. I’ve first-hand tested that census stats hypothesis in the middle of ugly disputes about “database geostubs”; you can learn very much indeed about somewhere based on the people who live there.
The real problem is when you go one step back. We have, thanks to Rambot, coverage of just about everywhere in the US. We’re reasonably near saturation in most developed English-speaking countries. This makes perfect sense; the nature of the English Wikipedia is that people who edit it will primarily be residents of developed English-speaking countries, and most interested in working on subjects of relevance to them. There are many fewer articles of relevance to parts of the world underrepresented amongst the project’s editor base. This is a natural emergent consequence.
But — as we know — natural emergent consequences aren’t always what people want to be the case. There is a strong argument for improving coverage of underrepresented parts of the world, markedly stronger than just about any other argument for writing about underrepresented subjects. (Our non-Anglophone coverage is awful. I can’t begin to impress on you how awful it is.) There are many ways to do that, but one of the obvious routes is…mass creation, isn’t it?
What was that you saw in the stubs? Villages in Central Asia? Moths in Madagascar? That’s not coincidence.
Many of the highest-profile mass creation incidents are explicitly based on an attempt at geographic diversity. Classically, mass creation in the last few years is an attempt to spam villages for developing countries the way we once did for developed ones. Developing countries, definitionally, have less opportunity to maximize their data integrity than their wealthier peers. This does not end well.
The famous case of the last couple years is Carlossuarez46, an admin who for many years mass-created location articles based on iffy data. Complaints about his work trace as early as 2014, but came to a hilt in 2021 when he made thousands of articles on Iranian abadis — a census classification that includes everything from towns to petrol stations. Attempts to untangle the resulting mess, where he accused everyone else of being racist for not wanting several thousand articles on Iranian petrol stations, saw him desysopped and declaring retirement from the project.
None of those articles would’ve existed if Suarez hadn’t mass-created them. Many of them don’t exist now, on account of they mostly aren’t actual towns and got deleted, but at least some of them were — and those wouldn’t exist, either. There are areas where mass-creation has produced mostly “real” towns; we have many stubs on villages in India and Turkey. None of those would’ve existed “independently”, either. There is no organic editor demand for these articles. There is demand for articles “about” India or Turkey or Iran, and this is how it’s filled.
Let’s trace back a little. Why is there a demand for nonspecific articles about India or Turkey or Iran, when there’s markedly less demand for articles on individual redlinked subjects from those areas? Most of this demand isn’t by people from India or Turkey or Iran, though we have such people on enwiki.2 Most of it is from Westerners who believe enwiki’s overrepresentation of the West is such a flaw that these articles are justified to stop it.
There’s a common response, when you talk about how mass creation burns the commons, which is: we need it anyway, because it’s a moral failure that we have articles on all these Western places and not these non-Western ones. There’s another version of that argument.
We no longer keep track of “who created the most articles”. The reason for this is Lugnuts. Lugnuts was a very bright, funny guy I always enjoyed interacting with, who is now permabanned by Wikipedia’s Supreme Court. These things happen.
Lugnuts is probably the guy who created the most articles. We don’t know for sure, because after he tried to kill himself over a notability argument we stopped tracking. It’s somewhere well past 90,000, though. A few of these articles are ones that passed Wikipedia’s peer-review processes (that’s its own post), but the overwhelming majority are mass-created stubs. Many were things like early silent films and — ahem — villages in Turkey, but a lot were people.
Lugnuts is a fan of cricket. Wikipedia has a lot of cricket fans. Cricket is a strange sport; it’s so difficult to make sense of as an outsider as to be a perpetual joke, and it’s a massive cultural juggernaut in a list of countries that does not include the United States. Like baseball, it’s a sport that lends itself to an analytical, stats-based, “nerdy” interest, meaning the Wikipedia sports editor base will tend to draw disproportionately from its aficionados. Accordingly, Wikipedia possesses both an array of lovingly-written cricket articles and a lot of confusion that they exist.
One thing that’s reasonably well-agreed-upon about sportspeople is that the media loves them. If someone has, say, competed at the Olympics, or played in the highest league of a reasonably popular game, they should certainly have coverage — “Hometown Hero” pieces in the local paper, if nothing else. Enough to write a Wikipedia article about. This works great, when you’re writing on the sort of scales people naturally write on. Any Olympian who competed since midcentury or so from a country with a reasonably developed press is going to have that coverage. Rather than making editors go out of their way to jump through hoops, someone interested in writing about a minor Olympian could rely on that heuristic and know they could write in confidence.
Lugnuts felt, even within this context, that we still didn’t have enough articles on Olympians and cricketers. He wanted to build the Sum of All Human Knowledge, and he wanted to build it on this alone. To this end, he scraped a couple of sports databases and used them to build, over the course of years, one-line stubs on as many Olympians and cricketers as he could hit.
The articles sparked uproars. People debated the ethics of mass-producing so many biographies. One particular sticky problem is determining whether people are living or dead; we assume people above 115 years of age are dead and below that are living, if we don’t have hard evidence either way, but we’d really not overrepresent supercentenarians that hard. Another, sometimes, is determining whether a given name is accurate — whether a Spaniard or Viet had diacritics stripped from his name that he used elsewhere, or whether two competitors with suspiciously similar ones were the same person through a database error.
But ultimately, the question was whether any of these articles would reasonably morph from stubs into fleshed-out biographies. The critical mass made it implausible. Another was the likelihood that anyone would stumble upon sourcing. When people make articles of their own accord, they make articles of people from times and places where they know how to find the sources. When they mass-create stubs, they don’t. Here came in the same problem as villages — systemic bias exists for a reason.
The argument wrapped around for years, sparked up on drama-boards regularly. Lugnuts sparred with deletionists who targeted him, sometimes rightfully (no database is ever perfectly clean), sometimes on the basis of his reputation. Eventually, he was sitebanned in a complex arbcom case (think “Supreme Court”) about the cesspool that was deletion-related arguments, which multiple people came out of with sanctions. He left a parting shot:
General consensus is he was bullshitting, trying to scare people at the last moment.3 Any representative sample of his articles doesn’t turn up the kind of subtle-trolling you’d expect if it were true. But it’s an explosion of agony and energy — someone who devoted his time to a project that no longer wanted him.
In the aftermath of the Lugnuts affair, people pushed through a change to Olympian notability. No longer was competing sufficient for an article — now, if you didn’t marshal the evidence beforehand, someone had to win to be presumed notable. This does not reflect any real underlying difference. There are no Olympians from a time and place with a developed mass media who don’t have significant coverage, no matter how they did. But when people intentionally create articles on people from times and places where that isn’t the case, things get weird.
This is not an apologia for deletionism. Deletionism is a pernicious overreaction, a death-grip on a murder weapon, the mad wishes of a tormented student. But it does not arise without motive.
Wikipedia has a hard time understanding that something can be neither mandatory nor forbidden. It is the last best place on the internet, a beautiful world beyond your comprehension, but God, I hate that shit so much — it’s the one romantic fatal flaw that will kill the place if anything does. Notability guidelines have a tendency to be misinterpreted as “the line at which there must be an article”, rather than “the line at which there can be”. People try to jack up standards higher and higher out of the terror of mass creation, which only cuts out viable articles and drives people away in disgust. There are plenty of contexts under which it’s possible to get rid of an article that technically passes notability guidelines but is ill-suited to a standalone — most famously in list form — but a notability-bashing misunderstanding of the inclusion threshold results in attempts to increase it alone, with little thought to solutions that lack collateral damage.
The necessary move is to target mass creation. Consensus has never exactly formed for anti-mass-creation sanctions, because people can never agree on their borders.4 But it’s the actual problem here, and the one completely invisible to readers outside the cost to Special:Random. Such articles produce a maintenance burden while providing negligible encyclopedic knowledge. Almost everyone agrees that we want to avoid them, but most discussion of how to avoid them falls into mandatory/forbidden fallacies, solving no problems while cultivating a reputation amongst outsiders for zealous deletionism. There is a solution to this, a way to cut the problem off at the root, but the world which admits it struggles to be born.
“Censuses” is ugly. “Censi” looks like you’re being wrong by accident. I’m being wrong on purpose.
Here is a subject that could have its own post, and perhaps its own book: the English Wikipedia is one of the few “abstract” (i.e. people meet people they don’t know irl, rather than people they do) parts of the internet that doesn’t massively underrepresent people from developing countries with many English speakers. In much of the internet, a native Anglospherian will meet more Scandinavians than Indians, despite this being wildly, hilariously misrepresentative.
There are still very few South Asians or West Africans on the English Wikipedia compared to their representation amongst English speakers, but it is a far smaller gap than most Westerners are normed on. In turn, the “core” editor base is still much more like most of the abstract internet; these editors tend to have a higher turnover and less involvement in backstage processes. There is a substantial “shadow Wikipedia” of articles almost exclusively read and edited by non-Western editors. Few of them are what you would call “good articles”. This is a constant bubbling-under discourse that no one really has.
"So, was this guy who just said some crazy shit after being sanctioned telling the truth?” is a complex question with an equally complex line of precedents. There’s another one that arose while I was partway through writing this post. We’ll get around to it.
This particular problem only arises from articles created at such a scale that a human can’t do it by hand. There are other ways to create a large number of articles. It’s possible to create one article a day over a medium-term horizon, though they will probably not be especially good. In shorter bursts, you can do higher. The individual most strongly associated with creating articles at this rate happens to write specifically about female scientists. You can imagine how these conversations tend to go.