What is Wikipedia? Wikipedia is a new paradigm in human discourse. It's a place where anyone with a browser can go, pick a subject that interests them, and without even logging in, start an argument. In fact, Wikipedia is the largest and most comprehensive collection of arguments in human history, incorporating spats and vendettas on subjects ranging from Suleiman the Magnificent to Dan the Automator. (links added)

— Lore Sjöberg, "The Wikipedia FAQK", Wired, 2006

Since its beginning, the English Wikipedia has used a consensus model: community discussions are the main process to implement, interpret, reinterpret and even form policies and guidelines. Over the years, the venues for this have grown and evolved. Currently, most of it takes place at one of a couple dozen "noticeboards", internal project pages in which threads are opened to address issues or open discussions. These range from broad discussions of core sitewide policy (hence why we call it Wikipedia:Village Pump) to conduct issues with individual users (hence why we call it Wikipedia:Great Dismal Swamp).

However, there is far too much of it for anyone to keep track of: since the beginning of 2025, there have been over six thousand threads on the noticeboards and village pumps.

Who has time for that?

Luckily for the person who wants to keep up anyway, most of these are somewhat inconsequential in the grand scheme of things (one person having a minor CSS issue on a specific skin, one person vandalizing a page and being blocked immediately). The more consequential threads are few and far between. But there is still an issue here: how can we distinguish between them? Even if 90% of threads are routine everyday issues, it is still quite time-consuming to go through a giant list and determine which 10% of thread titles will end up being a discussion of significant consequence.

Well, more significant threads tend to be longer. Often, the conversations with the most participants are those which examine Wikipedia's most interesting edges in editorial policy, coverage of content, and values of users. Discussions with high engagement are almost always conflicts and debates, where discussion participants are passionate about a topic and recruit others into the conversation. Noticeboard threads follow a power law distribution, and giving ourself a length-based cutoff sharply decreases the number of discussions to look at. But even then, hundreds of noticeboard archives would take days to go through and manually examine the section sizes.

This is where computerized analysis becomes useful. I wrote a program that would make the little electric person inside of the computer box look at every noticeboard thread after a start date, and compile a table of each discussion (its title, its URL, a count of its participants and its length) — this is what it had to say.

The above graph relates the number of discussions to the cutoff length, in tens of kilobytes. Roughly speaking, one character of plain text (with no formatting) is one byte, so a kilobyte is a thousand characters: the Rifleman's Creed is about one kilobyte, the short story The Gift of the Magi is eleven kilobytes, and the short story The Cask of Amontillado is thirteen. In Wikipedia discussions, there is a lot of formatting for bolding, underlining, italicization, links and templates: for example, my signature (jp×g🗯️) is 175 bytes but shows up as four letters and an emojus. Even the default signature — JPxG (talk) — is 44 characters of code for 11 characters of text. There are plenty of other situations where people use code in discussions, but every comment is guaranteed to have one signature, so we can assume right out the gate that the byte count of a discussion will virtually always be higher than the byte count of readable text.

Now, while I'm sure there is some way to determine an aggregate coefficient of discussion size to amount of rendered text, time constraints (as you will see) limit how much effort can be spent on this, so for the time being, we can somewhat approximately say that ten kilobytes is around the length of one Cask of Amontillado, or one Gift of the Magi.^[1]

What this means is that if we decide to only read discussions from 2025 of at least one cask's length, our task goes from reading six thousand threads to reading nine hundred: more than a six-fold reduction. And if we go to two casks, it becomes four hundred, and by the time we get to fifty kilobytes, there are only 118 threads from this year to date, which represent a pretty wide gamut of discursive events: those who have enough free time to follow these places on a regular basis will likely see a lot of familiar section headings.

This approach does, of course, give us some shortcomings:

• This analysis doesn't constitute an analysis of the whole edifice of discussion on Wikipedia. There are, for example, many discussions on individual talk pages of articles, as well as those for policies, guidelines, essays and process pages. Because of our rather haphazard development, there are actually discussions all over the place: there are some RfCs that happen on village pumps, some that are on policy pages, some on policy talk pages, some that have their own subpages, and some (almost all old ones) that are subpages of WP:RFC for some reason. There's also all of the stuff at WP:CENT, and arbitration cases, and AfD pages, and MfD pages (which sometimes set important policy precedents). In my mind, there swirl various ideas for an approach that will account for this lacuna, and allow a truly comprehensive analysis of all discussions that occurred during a given time period.
• A discussion being long doesn't mean it is important. It's possible, and indeed very common, for arguments with comparatively little importance to become humongous knockdown dragouts, simply because the participants are highly obstinate and refuse to drop the issue. Sometimes there is one participant who is highly verbose, and communicates via giant walls of text, even though the discussion isn't that big of a deal in the larger scheme of things. Sometimes there is a minor issue, but it can only be explained by copypasting a very long chunk of text or code or log output.
• A discussion being short doesn't mean it is unimportant. It's possible for a consensus to be reached quickly, or by a small number of people, that has far-reaching implications, especially if they aren't obvious at the time. It's also possible (although admittedly rare) that a gravitous discussion will occur where every participant's comments happen to be concise and brief.
• Some discussions are split up into multiple sections by use of "arbitrary break" headings, by virtue of technical necessity (large discussion sections lead to slow loading and edit conflicts). The present computerized approach makes it very difficult to detect these, instead seeing them as multiple smaller discussions.

Ultimately, however, I think that perfect is the enemy of the good, and these shortcomings do not eliminate the benefit of this procedure. The alternative to running a pre-winnowed analysis of noticeboard discussions is not an artisanal hand-crafted holistic analysis, but rather no analysis at all. Indeed, running this analysis in July gives a substantial backlog of discussions, even with a relatively high threshold, and time constraints would dictate an extremely sparse allotment of time to each. For the intrepid, there is also a truly massive, browser-groaning table of all 921 discussions above 10k.

One thing that's quite noticeable about these discussions is that many of them are very contentious arguments about user conduct issues. That is to say, they are "dramaboard" threads. This was somewhat unexpected; while I knew that there were a lot of these, and I knew that they got very long, I didn't think that they would actually constitute a majority of high-length noticeboard discussions. Perhaps this reflects negatively on us as a project — or perhaps it reflects negatively on a noticeboard-centric methodology for winnowing discussions. I think more analysis is necessary to figure out what's going on here. In the process of preparing this report, it was pointed out to me that this could lead down a dark path — those of a certain age may recall the heyday of Encyclopædia Dramatica with consternation.

It is true, I think, that including so many intensely-personal disputes in a list of most-participated-in discussions could end up being intrusive or even voyeuristic if done without sensitivity and care. Indeed, this is the same issue that occurs when writing the Signpost arbitration report — a column that often features lurid details of good editors at their worst. But noticeboard threads, like arbitration cases, bear heavily on the policies and guidelines of the project, and are indeed inseparable from them. Many important policies and precedents are based on specific incidents, and the same is even more true of our unwritten customs.

While we may have our personal disputes, we are ourselves the persons who shape the project, and this project remains a major participant in the online world's information ecosystem — many arbitration cases are central to our coverage on contentious hot-button issues, and obviously of great import to the project at large. For this reason, I think it is appropriate to include all noticeboard threads, even the dramaboards, and maybe even especially the dramaboards.

As a brief sample of what sorts of things this approach turns up — and again given the combination of time constraints with the large amount of time to be covered — I will give a table overview of noticeboard discussions above 50,000 characters closed between the beginning of 2025 and today.

Since this is a sortable wikitable, the way to view it sorted is to click on the top of the respective column: the default order has no particular significance.

In this table you can see a number of statistics for each discussion, aside from simple length. It's possible to count the number of comments in a discussion,^[2] and do a winnowing based on that, rather than simple volume of commentary. It's also possible to count the number of distinct signatures, which allows winnowing based on how broad participation was, rather than how much of it occurred. Furthermore, maximum indent level can be measured, which represents the longest exchange in a subthread. One may imagine other measurements, like average indent level, which would give an approximation of how much the conversation consisted of individual exchanges (e.g. a straightforward RfC where each comment was a response to the opening question would have a low average indent level, whereas a highly personal back-and-forth argument between individual users would have a high one, even if both had the same amount of text).

For example:

• Sorting by length, the biggest threads are an administrators' incident noticeboard thread about battleground behavior at capitalization RMs, a dispute resolution noticeboard thread about the Autism article, and a second AN/I thread about an individual user.
• Sorting by number of comments, it's the same AN/I thread at the top, but below it is instead a policy Village Pump discussion about LLM comments in project discussions, and an RfC on the reliable sources noticeboard about the Southern Poverty Law Center.
• Sorting by number of distinct users, it's the Village Pump LLM discussion, the Southern Poverty Law Center RfC, and a technical Village Pump discussion about the WMF's "Simple Summaries" project.
• Sorting by maximum indent level, it's almost entirely drama.

It is my great regret that I must leave you with a simple unrefined table of discussions, but vicissitudes in my own life have recently conspired to give me very little time for on-wiki activities. However, it is my plan to keep running this program for every issue.

Perhaps someone might step forward for upcoming issues to help summarize and analyze future batches!