Wikimania 2026 technology lightning talks: transcript

Key points

• Capacity Exchange lets Wikimedians find and message peers who share skills. It was built under Wikimedia Brazil coordination and has 1500 users and 52 organization profiles.
• Wiki Gender Stats replaces Humaniki, which stopped updating. It shows monthly gender gap data by project, profession, birth year and citizenship.
• C. Scott Ananian proposed a general annotation service anchored to wikitext ranges. It would support edit checks, suggestions, translation and Wikispeech without cluttering the markup.
• Reaction is a lightweight tool for small and medium sites that reads logs and runs commands to block abusive bots. A plugin system and a peer-to-peer cluster are in development.
• Tagzania is a Basque-language geo-guessing app with about 10,000 daily users. It uses Commons images, and players upload photos that are reviewed by humans before a bot sends them to Commons.

Chapters

1. 0:01Capacity Exchange peer-support platform
2. 4:26Wiki Gender Stats replaces Humaniki
3. 11:33Annotation service for MediaWiki content
4. 16:45Reaction: protecting against abusive bots
5. 21:54Tagzania geo-guessing game and Commons
6. 27:39Closing remarks

Full transcript

0:01I'm going to read my speech, because English is not my mother language, and also I'm pregnant, so I'm losing some words, so it's better for me to keep here on my notes. Okay, so you may have heard about Capacity Exchange, the CapEx, or perhaps this is your first time hearing about it. Either way, scan the QR code to visit capex.toolforge.org and log in using your Wikimedia Unified account.

0:36Before I introduce the platform itself, it's worth sharing a little bit about the story behind this project. So the capacity exchange project was born from a dream to address one of the largest missing pieces in the Wikimedia movement's toolbox, the peer support. Back in 2016, in the corridors of the Wikimedia movement's strategy discussions, a group of Wikimedias from different backgrounds dreamed of creating a sensually designed platform that could serve as a space where people could connect and make use of the collaborative and volunteer spirit that lies at the heart of the Wikimedia movement. At that time, it seemed distant but not impossible.

1:18After all, nothing is impossible for Wikimedians. And it proved to be accurate. We now have a fully functional, polished, and community-driven platform that enables Wikimedians worldwide to publish information about themselves and their affiliates, conduct searches, and connect directly with one another. But what is CAPEX, after all? CAPEX is an accessible, beautifully designed, multi-language, and user-friendly social-technical solution to look for and connect to your peer Wikimedians, who are willing to share skills, abilities, or services.

1:55Here you can see some of CAPAC's main features. Logging with your Wikimedia account through WWALF, using the platform in your preferred language, creating a profile to share your Wikimedia experience, showcasing your Wikimedia organization, and most importantly, finding and exploring people who are willing to share their knowledge based on language region affiliate or area of expertise once you find someone you'd like to connect with simply send them a message to the secure metal wiki messaging system and let the connection begin After a three-year development cycle, under Wikimedia Brazil Coordination, we now have a functional and rich platform.

2:40Our greatest challenge is to increase its visibility and ensure that it reaches the many communities, affiliates, and contributors across the Wikimedia movement. And that's where you and your local community comes in. If you would like us to introduce CAPEX to your community, please reach out to us by email or through our Telegram group. We will be happy to support your community in getting started. The story of CapEx is only beginning. Building a platform capable of facilitating meaningful connections was the first step.

3:13We believe the future of the project is bright and that it will continue to benefit people throughout the Wikimedia movement for years to come. But we need you to use it and make the most of what it has to offer. Throughout this project history, we had the opportunity to meet many remarkable people from across the Wikimedia movement. These interactions reinforced our belief that the platform like KAPAX can make it easier to connect the thousands of Wikimedias working in every corner of the world.

3:45Here are some of our supporters who have representatives on our advisory committee. Your community can be in this list too. We would love to have your input. It's also worth sharing that we currently have 1500 users and 52 organization profiles. So join us, create your user profile and start connecting with your peers on CapEx. The individual QR code available on each user profile is also a great alternative to sharing usernames on private algorithm-driven platforms when meeting people at conferences and other Wikimedia gatherings, as here in Wikimania.

4:26CAPAX was built by and for the Wikimedia movement. Explore it, use it, and let us know how we can continue improving it to better serve your community's needs. Thank you so much. Thank you very much. Okay, then the second presentation is about Eumanniki. Down here comes Wikigender stats and it will be Daniel who will present. Unfortunately, Natasha is a bit sick.

4:59okay so again five minutes here's the crowd thank you thank you i brought my emotional support Sure. I would do... What? Okay. so i'm here to present on behalf of lesson page i get some echo uh and you too uh to measure the gender gap on wikipedia and other wikis can you click Thank you. So why do we need a tool first? Because, well, first we have to prove that it actually exists, the gender gap actually exists. We all know that there is one, but it's better to be able to measure it with tools. We also want to track its evolution because, well, how do we know that we have an impact on the gender gap if we can't measure it?

5:53and we need to know where to focus our efforts to be more um what do we say i create to be more to have a better efficacy uh so uh historically there were a few tools that did that the first one was the uh i don't remember the name of the actual name uh it was an index to measure the gender gap specifically on wikipedia so we have some overview of what it looked like then around 2017 uh tools dedicated to the occupation gender gap which was named demilize was created and finally in 2021 the first tool was replaced by humaniki which you may have used maybe some of you which Show the gender gap for countries, wiki languages, and not occupations, sadly.

7:04Uh, I can't click. But last year, Humaniki stopped being updated and the less went offline. So we're kind of left without any tools. from this we decided to create a brand new tool which we called wiki gender stats and no wrong direction once again okay here we are so we combined both humaniki and which had a temporarily updated version, but only on the French Wikipedia.

7:52And we created this new tool, which I'm going to present right now. And I invite you to go take a look at yourself. So you can scan this QR code. It will bring you to the website. I have put some screen captures here in case it crashes. I already presented the tool earlier this year during the Wiki Women Summit. And the tool managed to stay online when several tens of people are connected to it. So, what does it do?

8:33You can see here, there are a lot of graphs everywhere, everywhere. The first one, you can see the gender gap on the various Wikipedias. You can select the project. Right now there is Wikipedia, Wikisource, and WikiQuotes in all their other languages. You can select the date. There is a capture of data every month, once a month. So we can track the evolution of gender gap throughout time. You can also see the gender gap by profession, right here, with a similar kind of graph, and the evolution of the gender gap through years, the birth years of every person on wiki data.

9:18And the one I'm most proud of is the gender gap depending on the citizenship of the people present in the wikis. For the colors, the yellow is women, the red is men, and there is purple for all other genders, queer identities, but as you can see... There's a nice name, so there's a bit of work here. So how does it work? Every month, all the human elements on Wikidata, which have at least one known gender, are all gathered on the tool's database, then all those statistics are pre-computed for each of the criterion, so date of birth, occupation, weekly language, and citizenship, and are accessible on the website and soon there will be a page where you will be able to make your

10:20own course on the database so you will be able to cross-reference the different criteria for instance if you want to know what's the gender gap for I don't know the people born between 1600 and 1700 in You will be able to do that to have more specific answers to your needs and to know where work is needed.

10:55Has it crushed it? Okay. Thank you. Thank you very much for your presentation. It's time to switch between the slides. So, well, actually, I forgot the run-off show, but amazing. Article in Notation by C. Scott. Do you want the clicker? Do you want the mic? I can do the clicker for you. Yes. Take that one. That one seems to work, too.

11:33Hello. And where should I stand where I'm not in front of it? The screen for everyone. Everyone So I can see. Great. Okay. Let's go. I'm Scott Mannion. I'm user cscott on the projects. And I'd like to talk to you briefly about the work I'm doing on annotations of media wiki content. Go ahead. Next slide. There you go. So what do I mean when I say annotations? Well, this is what typical markup looks like on the wiki mania pages and any translatable pages on multilingual projects.

12:03None of those highlighted clauses are content. their anchors and markers, trying to squeeze in some more information into the article that we've decided doesn't belong there immediately in line. In this case, translations into dozens of languages. Clearly, if you open the article and you saw that paragraph in 25 languages before you got to the paragraph two, that's not workable. This isn't really workable either. We're frantically trying to minimize what we put there, but it leads to very fragile and difficult to understand markup.

12:33Next slide. But there's actually lots of different types of annotations in our projects. In addition to translation regions, we do inline discussions with HTML syntax comments. We have various forms of citation and attempts to mark what regions those citations belong to, correspond to, or support, rather. for wikis with multi-script content we have markup to indicate how a particular word is transliterated into the various scripts that it has and wiki source uses different types of annotations we have lots of maintenance tags which are again trying to sort of put some process information in line in the article and hopefully a way which is non disruptive to editors but sometimes is sometimes isn't but in addition there's lots of other things that we'd like to do that have been stopped by the requirement of putting that information in the article somewhere. The biggest example probably is Wikispeech who comes to Wikimania sort of every year. So far I think editors have resisted putting pronunciations for hard-to-pronounce words in line in the

13:34articles on the grounds that it makes the articles harder to read which is reasonable but it prevents us from adding information to our projects. Which people are enthusiastic about editing. Next slide. So let's pivot for a moment. We have a problem on Wikipedia. You've heard, you'll hear from lots of talks about the difficulties people have editing our content, especially on mobile. And certainly seeing a page full of placeholder markup or weird annotations or even pronunciations doesn't help the issue. Next slide. So, a number of teams at the foundation are working on what we call structured editing initiatives to provide actionable ways to contribute, as illustrated here, prompts to specifically fix an issue to sort of get you over the blank page problem and get started.

14:18Next slide. So, these are two tools right now, edit check, which works as you type, sort of suggests things that maybe you could improve and suggestion mode which is more proactive says hey there's this article over here maybe you want to make an improvement to it right this this article is missing some citations or there's some other issue which you can fix next slide. So there's implementation there's some issues with the current implementation of edit checks the primary one is it's tied tightly to visual editor and visual editors internal data model It applies only to the latest revision so all those suggestions go away once your editing suggestion ends and it's hard to use them outside visual editor.

14:58Next slide. So. The proposal is to build an annotation service to fix these problems with the edit check. Next slide. And more than that, to provide a general annotation service that we can use to sort of unblock all these other things that people want to do on the platform. Fix translate, maybe allow wiki speech to add their annotations, things like that. Next slide. So we're getting, you know, I don't have a lot of time. This is the basic principles of the annotation service. Annotations are anchored to a wiki text range in a specific revision. They can get ported to more recent revisions or backwards to previous revisions.

15:35It's stable. It's explicitly migrated like it is for translate. When the text changes, the annotation may or may not apply, so there's a process to re-associate them. And it's mapped to HTML using Parsoid. So for people who are working with the HTML format on the page, including sort of readers on ReadView or mobile, you can map back and forth to the wiki text. And it's transparent, so there's no clutter in the actual markup itself. Next slide. So, the initial implementation is a data structure, some serialization with W3C standard formatting, some APIs, basic migrations on Parsoid, but the other thing is, hopefully, this is extensible.

16:15So, for you and your projects, if you've got a payload that would be useful for this, hopefully, we can talk about adding that. And... Concretely, next slide, sorry. Right now we're using for edit checks and suggestions. You'll see that towards the end of this year, early next year. Some of it's already in visual editor. We're hoping to use it for translate and sort of reader highlights next, and the future after that is up to you all. Next slide. And in my 10 seconds left, I'm going to wish you all a happy Brooke Vibber day. Yay!

16:45Yay! It's normally June 1st. I'm a little late, but I wanted to recognize her 26 years with the project. So clap for Brooke Weber. Yay! Thank you, Scott. OK, then moving on to the next one. So the next presentation is Towards Easier and Efficient Protection Again, Abuse Bots, Colombe. Reaction by Pomme. Thank you.

17:15Thank you. C'est les fleches? C'est les fleches. OK. OK, so hi, I'm here to present your action, which is a software made to help protect against abusive bots, notably AI bots that come in big waves on websites and hammer them and block the traffic. So if you know about this topic and your system admin, administrator, you may know about fail-to-ban. Fail-to-ban is an old software that's really used. It's packaged everywhere. It has lots of online resources. It's 25 years old, but it's really heavy on CPU and RAM, and it has a cumbersome configuration that has really bad documentation upstream.

18:02so i was first looking for alternatives but most of them are heavy software designed for big infrastructures or they are centralized uh they they work on a centralized database like crowd sec or they are very specific specific to one firewall specific to one type of website for example so when i started reaction i i tried my goal was to target small to medium infrastructures So it's not for Wikimedia scale, but it can be for independent projects, for smaller projects that depend on Wikimedia.

18:43The configuration was meant to be flexible, to start on a blank page, meaning there are no big defaults in the configuration. You start with your own configuration and it was made to be concise and possible to write in one file. So I also wanted good performance and for all those first goals, I can say that it's a success and I'm currently working on my last goal, more on that later. So Reaction reads the logs of the website, all the events that happen on the website.

19:22So it just launches a command that will follow the logs. then we can do some regex or json parsing on those event streams and we have support for both IPv4 and IPv6 with optional grouping on the CIDAR and then we will launch command as reactions to those events so when we see that, for example, one bot is looking at a lot of pages at the same time we can launch an action to block these bots and such and also it has a good good online doc which is an important feature to me so here's a minimal configuration example but I think I will just pass it because it's too technique it's it may be too technical But, yeah, we just have a command to follow the logs of a service. We have regexes to extract the IP from the bots in those events. And after a configured retries and period in which we are looking for IPs, we will launch a launch command to block and block those bots.

20:39Okay, so the core of reaction is mainly done. According to me, it's quite feature-complete and I'm working on a plugin system that can make reaction more complete and more performant, so that it can do lots of more things. So, I've already worked on firewall plugins to make your reaction faster, and I'm working on an HTTP malware to block IPs, so that if someone is blocked by error, the person will see a nice error message in the webpage, instead of just seeing nothing and the website don't load anymore.

21:24And I'm working on a peer-to-peer cluster so that the... So that multiple instances of reaction protecting multiple different websites can work together and share their bad bots so they can block IPs that were blocked from other instances. OK, so it's a package in most, in a lot of Linux distributions. And it's, thanks.

21:54And it has a dev file for Debian-based distribution. And OK, thank you. If you have any question, you can contact me on one of those protocols. And thanks for listening. Thank you. OK, so the last presentation will be by Luis. who is here just reading the titles that everyone will know it's probably already shown so yes uh taxania okay my name is luis joe fernandez and i'm presenting taxania which is a geoguessing game it's like geoguessing but in a small phone it's a mobile commercial app it's for profit yes it has ads and a premium version but it's in basque it's a minority language so it's kind kind of anti-commercial at the same time. It has 10,000 daily users.

22:56In a total Basque speaker population of 800,000 people. That's 10% of Paris population. Let's you guess. How do you play? We show two places. It's illustrated by several photos. I get 30 seconds extra. Thank you. You can place them in a map. It's an open street map in Basque. Basque photography all over the world. And you get points for guessing, proximity and time.

23:29faster, more points. So you have rankings, groups, sharing, and it's every day it's different for everyone. It's not like a Wardle or Anthropium or Maptop if you know that everyone gets the same challenge every day. Every player gets different challenges. It's done by a not mobile game company but by Python company. We work with Python and Plone and things like that. So the relationship with Wikimedia Commons the two places to get every day one the best country and the other one rest of the world are illustrated with free content which is of course Wikimedia Commons and when you get the results the page will show you the author the license and the link to Commons to get all the information of that particular picture but we also are generating user generated the content.

24:21Players can upload from a certain point, a point of interest, where they are placed, not gallery, only live from a certain point and you have to fill four pictures. In order to get to that, you have to go through several permission systems, it's well controlled, you have to allow a precise location. if you don't want okay you don't provide places and also there are explicit attempts of conditions of you have to check and they say that they mentioned the CCBASL license and also it's mentioned this will go to Wikipedia thank you very much and people checks then we process the uploads it's not AI on anything or automated by human with human checking we determine two things is it okay for Wikimedia Commons and is it okay for the game different

25:21questions but related the players get a little satisfaction they get a list of places they have uploaded if rejected they get the reason why it has been rejected and if approved they get a stats of how many people have seen that place and also the medium distance and and fastness of all the players And Wikimedia Commons gets a little satisfaction, too, because we are uploading the pictures and we created a Commons bot, asked for permissions, and it says users upload is dimmed OK for Wikimedia, a public URL is created, and four pictures are uploaded with each place.

26:08so so far it signs January 3,500 photos have been uploaded in the Basque country and the whole of the world a half of those pictures are in the Basque country they are concentrated there and that is the rest of the world the rest of the world for the Basques is very concentrated as you see but okay next if you scale if we do an internationalization it can be taxania France it can be taxania Camry it can be taxania Gujarat taxania Quebec we are open for partnerships what's the motivation for a possible taxania in another context. Well a big Texania can have revenue sharing and a little Texania, Texania Camry for instance for whales can have language advocacy and local

27:08comments content it's a shared motivation. So, if anyone is interested, it can be managed by a person, we think. The workload can be from little to very much and we are getting some revenue with 15,000 monthly users and we have also premium things and that's all. If you are interested in contacting, take a picture.

27:39no QR code I am allergic that's it. Thank you very much so it was the last presentation for the lightning talk showcase for technology thank many thanks everyone for attending I'd like to thank also Valdir who was the room manager who helped for previous sessions

Source and licence

Recording: File:Wikimania 2026 – Technology Lightning talk series.webm by Amanda Jurno, Joris Darlington Quarshie, Danÿa, Luistxo, Natacha Rault, ppom, C. Scott Ananian (2026-07-24), via Wikimedia Commons. Licence: CC BY-SA 4.0. The transcript, summary and chapters on this page are shared under the same licence (CC BY-SA 4.0).

The transcript is machine speech-to-text by the TidyTools Audio & video transcriber; the summary, key points and chapters were written by Claude from that transcript. Both may contain errors: check the recording before quoting it.