The Signpost


Technology report

Community Wishlist rebooted, a new pie chart tool, and a whole lot about citations

A Wikipedian approaches the oracle

An updated plan for the Community Wishlist

[edit source]

For years, the Community Tech team was dedicated to managing and implementing the Community Wishlist, before the WMF decided to disband the team and turn it into a program. The Wikimedia Foundation made the decision as the engagement with the Wishlist had sharply dropped in recent years, after the program had been restructured in 2024. As a result of the Tech team's disbandment, discourse erupted over the decision. The community raised concern over the loss of a major communication venue with Foundation staff, and its associated layoff of Tech team staff, some of whom were heavily involved in developing the Wiki Workers United union. More information on the disbandment can be found in our June technology report.

The new wishlist process is to be divided into four stages, those being Triage, Voting, Post-vote, and Implementation. A more detailed summary of the process is available over on the page about the updated process, with more information to come on the Community Wishlist updates page.

Triage

[edit source]
  • Submissions for wishes are already open, with the submission period to close sometime between November and mid January, depending on what works logistically.
  • A working group composed of community member and Foundation staff will be established. (look out for post about how volunteer engagement will occur) The working group will aim to have a diverse representation from across different Wikipedias and its sister projects.
  • The working group will be activated annually for four weeks to triage wishes prior to the vote, but does not include futher planning for how the wishes would be implemented. This year, triaging will be conducted in November.

Voting

[edit source]

The first voting period is currently expected to be in January, running from the 8th to the 24th.

  • Users will vote for individual wishes using a "Support wish" button.
  • Opposition of a wish may be discussed on its respective Talk page.
  • Users can vote for any amount of wishes, but cannot vote the same wish multiple times.

Post-vote

[edit source]

After the voting period closes, wishes will be sorted by their vote count. From here, the Foundation will decide on the wishes they plan to work on.

  • There will be three "project families" that wishes will be sorted into, with those being "Large Wikipedias or covering all wikis", "Small and medium-sized Wikipedias" and "Sister projects".
  • The votes will only be a guideline, and the Foundation may deviate from them where they judge a small-project wish as able to have a large impact, where there are not enough resources to work on the wish, where multiple highly voted wishes fall under the scope of the same team, or where multiple smaller wishes can have a greater impact than fewer, larger ones.
  • Where the prioritized wishes do not line up with the most voted ones, the Foundation will aim to "provide transparent reasoning prior to implementation", and the working group will be consulted.
  • Prior to starting the work, the community will be asked for feedback on the final list.

Implementation

[edit source]
  • The implementation of the wishes will be handled through the Wishlist Program and by a Lead Program Manager.
  • Submissions for the next vote will open once they have prioritized the current wishes.
  • The Foundation may engage in research and testing for wishes targeted at those typically not engaged in the Community Wishlist discussions, such as readers and new editors, prior to further development on the wish. – M

Journal Cited by Wikipedia compilation now reports open access status

[edit source]
A drastic rise in the percentage of DOIs flagged as free-to-read can be seen in September 2026

Since August of this year, the Journal Cited by Wikipedia compilation reports whether journals are open accessible, and whether publishers publish open access journals. The feature is fresh, so tweaks are likely to happen, but what is currently there is very functional. Essentially, JL-Bot looks at various categories present on an article and determines as best it can if a journal/publisher can be classified according to the table below:

OA
Open : Open access publisher/publication. This includes CC-licensed journals, Diamond open access, Gold open access, etc.
S2Open : Subscribe to Open publisher/publication
Delayed : Delayed open access publisher/publication
Hybrid : Hybrid open-access journal publisher/publication
Closed : Closed access publisher/publication
Unflagged : Bot finds no information. For publishers, this likely means the publisher does not have a consistent open access publishing policy, and thus cannot be categorised. Or it could be that no one bothered to check.


An example of where this can be seen in action is Wikipedia:WikiProject Academic Journals/Journals cited by Wikipedia/Target1. The table is sortable, so you can easily browse targets by open-access categories.

In light of this more-easily findable information about which journals/publishers are openly accessible, a parallel effort by myself and others to tag citations with |doi-access= was happening. In August 2026, there were about 27.99% of DOIs that were tagged as free to read. This percentage has remained relatively constant, in the 27% range, never reaching 28%, since the compilation started tracking |doi-access=free usage (see chart on the right). In September, usage increased by a massive 1.25% to reach 29.24%. – H

New pie chart tool

[edit source]

A new extension has been enabled on Commons allowing users to easily create and modify charts from a dataset. On pages in the Data namespace whose name ends with .chart, a new Edit with form mode can be selected from the top right of the page. This will bring up a menu, allowing editors to select a data source, the chart type, display title, and other associated characteristics. Users will also be able to see a live preview of their chart as well as the generated code. – M

Retraction Tracker for Wikipedia

[edit source]

A coder known as ThinkingPark has designed an LLM-based Retraction Tracker for Wikipedia. According to the documentation, the tracker cross-checks citations against the Retraction Watch retraction database (see previous Signpost coverage) to determine if a citation is retracted, then tries to determine whether the supported claim is presented as true, or if there are disclosure that the claim has been retracted in the main text of the article. No date is associated with the project, but the fix tracker goes back to 6 October 2026.

Specifically, the tracker operates like this:

  1. Get the list of retracted papers (including the reasons for retraction)
  2. Find the Wikipedia articles that cite them (through their DOIs and PMIDs)
  3. Find the exact passages supported by the retracted papers
  4. An LLM model reads the Wikipedia article
  5. An LLM model compares the passages and determines if it needs additional disclosure, correction, or if readers have been told the supporting paper was retracted

Every passage gets a priority based on how bad the reason for retraction is and how often the article is read. This way a "low severity" claim read 20000 times a day may be considered a higher priority to fix than a "high severity" claim read only 20 times a day. No edits are made to Wikipedia, the tool only tracks these issues on a daily basis. The documentation contains all the warnings you'd expect to encourage responsible use of the tool, aka ‘this is an LLM-based tool, it's not perfect, use your brain’.

As of writing (15 November 2026), the tracker finds

  • 65 "claims to fix" in 43 wiki articles
  • 64 "Sources to replace" passages in 36 wiki articles
  • 590 "Already covered" passages in 280 wiki articles
  • 29 "Fixed" passages in 10 wiki articles

Hopefully this tool can help clear our retraction-related backlog! – H

Citation bot clears its bug backlog

[edit source]

On 7 November 2026, something strange and unusual, something almost spooky, happened. Citation bot had cleared its backlog of bug reports. Contrast with the situation in February with over 100 active discussions (not all of them bugs). All in all, there have been over 271 closed discussions (which covers not only bugs, but also feature requests, general improvements, recognizing free DOIs like in the above story, and general questions about the bot) since the start of the year as of writing. This is a Herculean effort from primary bot maintainers AManWithNoPlan and Redalert2fan, but also from the community at large, identifying bot issues and making suggestions for improvements.

Is the bot now perfect? The answer is still a no (see previous Signpost coverage). The core behavior is as solid as it’s ever been, but there will, inevitably, always be corner cases and bad metadata. – H

Signpost
In this issue
+ Add a comment

Discuss this story

To follow comments, add the page to your watchlist. If your comment has not appeared here, you can try purging the cache.
No comments yet. Yours could be the first!





















Wikipedia:Wikipedia Signpost/Next_issue/Technology_report