The Team That Granted Wishes

What the Community Wishlist gave Wikipedia’s editors, tool by tool, and what it meant to be heard.

Photo by abigail low on Unsplash

For nearly a decade, Wikipedia ran a wishing well. Every year, editors from hundreds of language communities lined up to ask for things: a way to monitor a page for just thirty days and then let it go, a tool that could reveal who wrote a particular sentence, a system that caught plagiarism before readers did.

A small team at the Wikimedia Foundation, whose mascot was a dog wearing a Santa hat, took the most-voted wishes and built them. Watchlist Expiry had been requested since at least 2006. They shipped it. Who Wrote That? was considered one of the hardest asks ever to come out of a survey. They shipped that too.

In May, the Foundation dissolved the team, with wishes still on its workbench. To understand what that decision actually cost, you have to know what the wishing well gave, tool by tool, and who reaches into empty air now.

The ritual

It began in November 2015, borrowed with admiration from Wikimedia Deutschland, whose Technical Wishes team had been polling German-speaking editors since 2013.

The Foundation’s new Community Tech team scaled the idea to every project and every language: two weeks for editors to propose anything they wanted, in any language they wanted, with volunteers translating so a wish written in Hungarian could be weighed by an editor in Indonesia. Then two weeks to vote.

The first survey drew 634 people and 107 ideas, and the team made a promise that sounded almost reckless for a five-person crew: the top ten vote-getters became their backlog. Not suggestions to weigh. The backlog. The team later admitted the whole thing had been “a big, scary experiment.” It worked anyway, which is how most Wikipedia initiatives are born.

Anyone who lived through a survey season remembers the feel of it. Wishes went up in December like letters mailed north, which is how a dog in a Santa hat became the team’s emblem. Editors campaigned for their proposals on village pumps and talk pages. Proposers watched their wishes climb or stall the way other people watch election returns.

The categories read like a map of everything unglamorous and essential about the project: admins and patrollers, anti-harassment, citations, multimedia, bots and gadgets. By the 2023 edition, the ritual had grown to 212 proposals from 1,439 contributors, who cast 8,492 support votes.

The voting was never a pure popularity contest, and that was part of its integrity. Oppose votes counted as intelligence.

In the first survey, a proposal to let editors keep a watchlist of other users drew strong support and anguished opposition. The team listened to both sides and declined it, on the grounds that a tool for following people is a tool for stalking them. A wishing well that grants every wish is a hazard. This one had a conscience.

The thrill of delivery

Then there was the part few expect from institutional software development: the wishes came true.

The very first survey’s top wish, with 111 votes, asked someone to do something about link rot, the slow decay of the citations that hold the encyclopedia’s credibility together. Every dead link is a small betrayal of the reader, who clicks footnote 12 to check a claim, lands on a 404 page, and in that instant watches a verified fact become an unverifiable one. Editors had been repairing these by hand for years, a losing race against the entropy of the web.

The race was even more lopsided than it looked: a 2024 Pew Research study found that a quarter of all webpages that existed at any point in the previous decade had already vanished, and that more than half of English Wikipedia’s articles carried at least one dead link in their references.

The work that followed the wish, in collaboration with the Internet Archive, turned dead-link repair into industrial-scale rescue. The Archive began capturing nearly every URL cited across some 300 Wikipedia sites as the citations were added, at a rate of about twenty million URLs a week, so that a lifeboat would be waiting before the link ever sank.

IABot then swept the wikis swapping archived snapshots for broken citations, and the count tells a story of acceleration: nine million links rescued by 2018, more than fifteen million by early 2023, and past thirty million across hundreds of wikis as of this spring, andstill climbing.

The Archive’s researchers reported this April that on the web at large, falling back to archived copies cuts the share of unreachable pages from one in four down to one in ten. It is one of the great unsung victories of the open web, and it started as a wish.

CopyPatrol arrived in 2016, a clean interface wrapped around a volunteer’s plagiarism-detection bot. Whenever an edit added a substantial block of text to an article, the bot checked it against a professional plagiarism database, and CopyPatrol laid the new text beside its suspected source, so a reviewer could see at a glance whether someone had pasted in a newspaper paragraph. From there the workflow was three moves: revert the edit, leave a note for the contributor, close the case.

The stakes were higher than they sound. Copyright violations are one of the few problems that can hurt Wikipedia in a courtroom as well as in a reader’s trust, and before the tool, patrol meant deciphering raw reports, a grim backlog tended by a few stalwarts. After it, the backlog cleared and new patrollers showed up, because the tool made the work efficient. The team’s best work lowered the cost of caring.

Who Wrote That?, the fourth wish of 2017, was the kind of request engineers wince at. Editors wanted to hover over any sentence in an article and learn who wrote it and when, without spelunking through thousands of revisions. Tracing one dubious sentence through an article’s history meant searching the archive by hand, opening diff after diff, sometimes for an hour, to answer a thirty-second question: who put this here?

The answer matters constantly. It is how editors spot a slow pattern of promotional edits, catch a sockpuppet’s fingerprints across articles, find the one colleague who can explain a strange claim sourced to an offline book, and tell a typo from a campaign.

The team called this one of the hardest projects the survey ever produced. They built it anyway, as a browser extension riding on the WikiWho service, and turned the article history from an archive into an x-ray.

SVG Translate is a resurrection story. Wikimedia Commons holds hundreds of thousands of diagrams with text baked into them: the labeled heart, the water cycle, the anatomy of the ear. Translating one used to mean downloading the file, installing vector-graphics software or hand-editing the file’s source code, replacing each label without breaking the layout, and uploading the result. In practice, most diagrams never got translated at all.

A tool that solved this had existed once and died in 2009 when its maintainer drifted away. For nearly a decade, the ninth wish of 2017 sat there describing the loss. The team rebuilt it from scratch as a simple form: the labels on one side, empty fields for a new language on the other, one button to publish. Hundreds of language communities got their diagrams back, and a biology classroom in Dar es Salaam got a heart that speaks Swahili.

Event Metrics, the third wish of 2017, served the movement’s outreach wing: the librarians, museum staff, and campaign organizers who run edit-a-thons and need to show a grantmaker or a partner institution what an afternoon of editing produced. Before it, that meant counting contributions by hand or begging a technical friend for a script. After it, an organizer could define an event, list its participants, and get back the articles created and the words added to show impact immeditely

When it shipped, the Community Tech team did something no roadmap requires: they pinged every editor who had voted for the wish, one by one, to say it was live. Someone asked for something, and a year later the builders came back to that person’s talk page to say it existed.

Watchlist Expiry, the seventh wish of 2019, sounds small only to someone who has never kept a watchlist: the editor’s nervous system, the feed of every change to every page that editor has promised to monitor. The trouble was that the promise used to be permanent for every page. An editor watched a page during a deletion debate or a vandal wave or a week of protection, and it haunted the feed for the next decade, until veteran editors carried watchlists thousands of pages deep, mostly dead weight, burying the changes that needed eyes.

The wish asked for a clock: watch this page for a week, a month, or a year, and then be released. It had been requested since at least 2006. The engineering was, by the team’s own account, large and complex, and they spent much of 2020 on it, all so that an editor’s attention, the scarcest resource in the whole system, could finally be temporary.

Convenience wishes earned their keep too, because convenience compounds across millions of edits. Realtime Preview, the fourth wish of 2021, let wikitext editors watch their code render as they typed instead of cycling endlessly between preview and publish.

From 2015 through 2020, thirty top-ten wishes were completed. After the team rebuilt its prioritization in 2021, forty more were finished in three years, ten by Community Tech itself and thirty by other Foundation teams the wishlist had learned to enlist. The team even invented the Wishathon, a quarterly internal hackathon where staff from across the Foundation spent a week granting extra wishes, like cooks coming out of the kitchen to make whatever the room wanted.

The wish that outgrew the well

A fair question would be whether the wishlist was just power users voting for power tools, a survey of the one percent optimizing for itself. Dark mode is a good answer to that skepticism.

Editors had been asking for it since at least 2019, when it ranked second in the survey. It came back year after year, always near the top, always judged too large for a small team to attempt. The demand was so persistent that a few staff members built a dark mode script as a side experiment, and it spread from wiki to wiki as a popular gadget. In 2023, dark mode topped the survey outright. One voter’s entire pitch: “This could be the year!”

It was. The wish was bigger than the team, so the wishlist did what it had learned to do with big wishes: it made the demand undeniable and handed it to the Web team, which spent a year on it and shipped dark mode in July 2024 to everyone, readers included. Volunteer template editors repaired pages everywhere so articles would render cleanly in the dark.

The Foundation’s own analytics showed that one in five readers had already set a dark mode preference at the operating system level. A fifth of the reading public, billions of pageviews worth, had been asking for this in the only language their devices could speak, and the editors heard it years before the analytics did.

What it meant

The tools by themselves do not capture what the wishlist was. Wikipedia’s editors occupy a strange position in the world. They build one of the most consulted artifacts in human history, and almost nothing in their daily experience reflects that. No paycheck, no title, and most days no acknowledgment beyond the work itself. The relationship with the Foundation that stewards their platform has often run one way: announcements travel down, feedback travels up, and somewhere in the middle it evaporates.

The wishlist reversed the current. Once a year, the people who write the encyclopedia told the people with the engineering budget what to do, and the engineers did it. A crack team existed whose entire mandate was the editors themselves: their stuck workflows, their decade-old annoyances, their tools held together by one volunteer’s spare nights and weekends.

The team had names editors knew, listed at the bottom of each survey like film credits. They answered on talk pages. They published status reports admitting when something was harder than expected. They sent a newsletter. When they declined a wish, they explained why, in public, and the explanation usually held up.

The trade ran deeper than it looked. Companies spend fortunes trying to learn what the wishlist delivered to the Foundation for free: hundreds of feature ideas a year from the most knowledgeable users on earth, already prioritized by vote, already translated across dozens of languages, already stress-tested in public argument, each one arriving with the workflow context that user research teams spend months excavating. The editors gave their ideas as gifts. The team returned those ideas realized, as gifts in the other direction.

That loop, run annually for a decade, is how a Product and Technology department comes to know its audience by name, by history, and by the particular shape of each request or complaint.

That is what validation looks like in a volunteer community: responsiveness, which is rarer than praise and worth more. The wishlist was a standing institutional admission that the editors’ time mattered enough to spend engineers on, and every shipped wish renewed it. An editor could disagree with the Foundation about nearly everything and still feel, each December, that one well ran clean.

The severing

The annual survey held its last edition in 2023. What replaced it was announced in reasonable language: a continuous intake system for technical requests, better prioritization, better resourcing, better communication. Always open instead of once a year. Routed to the right team instead of piled on one. On paper, an upgrade.

And the people running it worked in good faith; the first community Wishathon, in 2024, came from that era.

But something had changed in kind, and editors said so at the time. When the Foundation paused the 2024 survey to build the new system, one participant wrote that the problem was “long grave neglect” of actually building what editors asked for.

Another warned the team that “antagonizing and not listening to the very community whose input you sought” was a reliable way to make the new wishlist fail.

Because a wish submitted to an open queue is a ticket. It enters a prioritization framework, receives a status, and waits its turn against everything else on an agenda the community no longer sets.

The December rhythm died, and with it the campaign season, the scoreboard, the translated arguments, the nightly refresh of the vote tallies, and above all the promise: that the top of the list became the work, period.

The first team had sworn, in effect, that the community’s voice would never be merely an input to a roadmap. The new system made it exactly that, and the difference between a covenant and a suggestion box is the difference between being heard and being processed.

Every shipped wish was a deposit in an account the Foundation drew on whenever it needed the community’s patience, and the ritual was how the account stayed funded. Goodwill of that kind is renewed annually or it decays. When the exchange of gifts became a queue of tickets, the deposits stopped, and the Foundation began spending down a balance it may not have known it had.

Then came this May, when the Foundation dissolved the community team itself and announced that its remaining work would be “divided and incorporated into existing roadmaps.” The wishes survived the wishing well, but only as line items, distributed among teams whose agendas they will now compete with rather than define.

It is the precise institutional opposite of a dog in a Santa hat.

The balance came due within days. An administrator, Tamzin Hadasa Kelly, opened a petition whose signatories pledged to join an editorial strike if the Foundation’s nascent staff union called one, and more than one thousand editors signed it. One wrote that the wishlist had been “going in a really bad direction for a couple years now,” and the dissolution confirmed the trajectory.

Jimmy Wales came to the village pump to reassure the community that it was “time to get serious about meeting community needs,” and the room heard the opposite of reassurance. One editor called the attempt to frame the dissolution as progress “infuriating.” On Commons, a veteran observed that the Foundation’s decision-makers did not seem to grasp “just how poorly the editor community regards them.”

Wikipedians, as a rule, do not strike. That hundreds were signing their names to the possibility is the most available measure of what the wishing well had been holding in place: trust.

A wishing well only works if someone is down there, waiting, catching what falls. For ten years, someone was. The coins are still coming. The well is still there. What ended was the listening.