Where Wikipedia fights

Wikipedia has no editor-in-chief and no final say. When two people disagree about an article, the argument ends the way it does in a word processor: someone hits undo. That undo has a name — a revert — and every one is logged and tagged, which turns a normally invisible thing into a countable one: how much of an article’s writing gets thrown away. Take Wikipedia’s own list of controversial articles, read the most recent five hundred edits of each, and you can see where the reverting runs hottest. It is not quite where the shouting would point.

The undo button

Reverting is Wikipedia’s immune system. The site’s founding rule is that articles take a neutral point of view, and its founding enforcement move is the revert: any editor can roll a page back to an earlier version with a click, and because the software keeps every version forever, nothing is ever quite lost. Most reverts are dull and uncontested — a bot catching vandalism, someone undoing a broken link. But when a disagreement is real, reverts begin to answer one another, and the page enters the loop every Wikipedian knows: edit, revert, re-revert, until an editor trips the three-revert line that gets an account blocked or an administrator locks the page outright. What the revert log records, then, is not opinion but friction — the exact points where the encyclopedia’s consensus machine grinds instead of turns. Counting it is about the closest you can get to a map of where writing the reference book actually hurts.

The most-reverted pages

At the top sits East Germany, where about 41.6% of its last five hundred edits were undone — better than one in three. Nutrition and Eshay are close behind. The list that produces is a strange one: a few articles you would expect, and several no one would file under controversial — a piece of slang, a stretch of tax law, a page about a single calendar year. The median article on the whole list has only about 9.8% of its edits reverted, so the leaders sit far out on their own.

Two kinds of article crowd that top. There are the ones you would guess — long arguments over history, science with a political charge, the framing of a country or a cause — where the reverting is the residue of a real fight. And there are the ones you would not: low-traffic pages that happen to be magnets for the bored and the mischievous, where a high revert rate is mostly the sound of vandalism being swept back up. The rate alone cannot tell the two apart. That is the next chart’s job.

Show the data

Article	Edits reverted (%)
East Germany	41.6
Nutrition	40.2
Eshay	37.8
Entertainment Softw…	35.2
Sex differences in …	34.6
Tax	33
Political correctne…	32.2
Radical right (Unit…	31.6
2012	31.4
American Airlines F…	31.2
Discrimination	30
Disc jockey	29.4

Source: MediaWiki revision history; counted in scripts/fetch-wikipedia.mjs

Two kinds of busy

A high revert rate hides two different fights. Some of these pages barely move — a handful of edits a month, most of them drive-by vandalism that a bot or a watchful editor quietly undoes. The rate is high because the traffic is low and the noise is bad, not because anyone is at war. Others are edited constantly. Sort the same articles by edits per day and a nearly different list appears: Autism, Korean War, and Artificial intelligence lead it, pages that change hands many times a day.

Show the data

Article	Edits / day
Autism	1.6
Korean War	1.4
Artificial intellig…	1.1
East Germany	0.9
Sigmund Freud	0.9
Poland	0.9
Canada	0.9
Anarcho-capitalism	0.9
Ku Klux Klan	0.8
Genocide	0.8

Source: MediaWiki revision history

Most of that churn is not reverted at all — it is collaboration, people building an article faster than they undo one another. The genuine edit war is the rare page that scores high on both counts: East Germany is one, heavily reverted and heavily edited, a running battle rather than a cleanup. But it is the exception. The most-reverted and most-edited lists barely overlap, which means the single word contested is really covering two conditions — a page nobody will leave alone, and a page nobody can agree on — that happen to look alike from a distance.

Where the fights cluster

Group the articles by the subject headings Wikipedia files them under and the spread is narrower than the noise would suggest: every subject lands between roughly 8.3% and 15.4% reverted. The order, though, is a mild surprise. The heaviest reverting is not in politics or science, which sit squarely in the middle. It is in Languages, followed by law, media, and questions of sex and gender — the arguments about words, categories, and what to call things. The calmest subject is Environment.

Show the data

Subject	Edits reverted (%)
Languages	15.4
Law	14.8
Media	14.6
Sex & gender	14.3
History	14.1
Religion	13.6
Politics	13.3
Science	13.3
Psychiatry	11
Philosophy	10.9
Entertainment	9.9
Technology	8.8
Environment	8.3

Source: MediaWiki revision history

The pattern is faint, and the samples inside each subject are small, so it is a lean rather than a law. But it leans the same way the revert button does. Facts can be sourced and settled; a dispute over a citation ends when someone produces the citation. Definitions cannot be sourced the same way, because the disagreement is the point — and those are the edits that get undone, restored, and undone again.

The quiet list

It is worth saying how quiet most of even this list is. These are the articles Wikipedia’s own editors singled out as contentious, and still the typical one keeps nine in ten of its edits, and only 6 of the 178 sampled are locked against ordinary editing at all. Locking a page is the administrators’ last resort, and how rarely these need it says something in itself: the three-revert rule and a large enough crowd of watchers settle most fights before they ever harden into the kind that needs a lock.

The fighting is real, but it is concentrated. A small number of pages absorb almost all of it, and the encyclopedia around them is a good deal calmer than its reputation for chaos would suggest — which is, when you think about it, the more remarkable outcome. A quarter-billion edits a year, no one in charge, and the median contested article still ends each day a little better than it started.