Wikimedia Apps/Team/Android/Anti Vandalism

Background

The Android team has received requests from various language communities to improve the quality of edits made through the app. Community members have asked for moderation tools that are fair and unbiased across language wikis. The team aims to enhance existing workflows while ensuring equity in moderation, an example can be found in the 2022 wishlist.

With recent updates to the app, such as the native watchlist, contribution history, and edit history, as well as the addition of the undo and rollback button on the diff screen, there is an opportunity to create a moderating solution within the app. The solution will build on existing mobile micro contribution workflows.

The team plans to develop an anti-vandalism tool (V1) that caters to experienced moderators across languages. Additionally, they will introduce a training task to help experienced editors learn patrolling if they haven't done it before. This will be done in close collaboration with the community. The team will also work with the research team to improve tools like ORES and collaborate with the Moderator tools team, who are addressing this issue on Mobile Web, to ensure a seamless experience for cross-platform users.

Product Requirements:

• Editors can review diffs for a variety of edits
• Editors have the option to 'revert' or 'skip' suggested changes
• Recent edits made in various languages can be displayed
• Editors can create and save warning messages to use when reverting changes
• Experienced editors will be introduced to patrolling on the app in a later version (V2)

Technical Limitation

Currently, it's not possible to have multiple languages in a single feed, but users will have the ability to switch between languages.

Hypothesis & Validation

Hypothesis & Assumptions

We believe that simplifying the patrolling tasks and structuring them as advanced-level tasks would be beneficial. Experienced editors without sysops or rollback rights could be assigned easier patrolling tasks or a training task, and once they reach a threshold of initial patrolling activities, they can be promoted to higher levels. To achieve this, we plan to start with a proof of concept for general patrolling of edits based on users with rollback rights.

We understand that patrolling recent edits is a priority and more suitable for mobile than tasks like NPP (New Page Patrol), AFD (Articles for Deletion), and AFC (Articles for Creation). Additionally, we expect that there will be community consensus on implementing a training task across different languages.

Validation

• Key Indicator 1: 65% of target mature audiences using the tool find it helpful for maintaining the quality of wikis and would recommend it to other patrollers
• Key Indicator 2: Increase in edits made by mature audiences by 5%
• Key Indicator 3: 10% of target mature audiences engage with filter preferences
• Key Indicator 4: 65% of target mature audiences engage with the tool at least three times in a thirty-day window
• Guardrail: Experienced users without rollback rights, both those who have and have not used alternative patrolling tools, should equally understand the workflow
• Guardrail: There should be no reports of the tool being used to negatively target underrepresented content or contributors based on in-app reporting mechanisms

Things that would be helpful to know

•   How does use of our tool compare to other patrolling tools when looking at MediaWiki Tags (SWViewer, Huggle, and Twinkle)
•   Do we see an increase in Undo/Rollback/Thank events
•   How popular is this task with our target audience relative to other Suggested Edits task?
•   What actions are most popular in the feature?

Risk management

Risk	Cause	Level	Response	Response Action	Trigger	Contingency Plan
Tool causes a decrease in new editor retention on wikis where it's available	Editors use the tool to increase the volume of edits they're able to revert, but do not leave talk page messages to explain the revert, leaving new editors confused and unmotivated.	Medium	Monitor	Measure the retention rate of editors who had a revert from the patroller task. If they got a talk message, did that increase their retention?	Qualitative complaints	If a user has not left a talk page message after using undo or rollback and attempts to move to next edit, add a prompt asking/requiring them to leave a Talk Page message
Policy disagreements between existing patrollers and newer patrollers using Edit Patrol / New patrollers use the tool a way that confuses other patrollers. Behavior and/or communication from App patrollers is seen as insufficient or lacking by other patrollers.	A new group of editors are able to start patrolling at scale using Edit Patrol, but they are not aware of existing policies or norms.	Medium	Mitigate	Post updates, and invite to office hours patrolling groups (CVU, recent changes patrol, small wiki monitoring team, stewards, global sysops, global rollbackers) before full release. Share Example Messages with communities.	Qualitative complaints from patrollers	Add more policy and template onboarding to tool, and revisit Example Messages after Community Configuration 2.0 is released. Continue work on building training task.
An increase in edit wars	Patrollers reviewing the same feed of edits go back and forth reverting each other's changes because revisions or rollbacks made in the tool appear in the feed for the next person (if no filters are applied).	Low	Accept	-	Qualitative complaints	Add more onboarding about policies surrounding Edit Wars. Add in-app warning for edit-war behavior.
The tool is being used to negatively target underrepresented content or contributors.	Patrollers are systematically reverting good faith edits for particular users or languages, harassing, or mistreating contributors through the tool.	Low	Mitigate	Unless otherwise requested per wiki, the tool available to those with rollback rights.	Qualitative complaints, in-app reporting mechanisms, guardrail analysis by Product Analyst	Add universal code of conduct onboarding to tool.
The tool causes an increase in vandalism, and increase in amount of work for existing patrollers	Vandals use the streamlined interface to quickly revert or rollback large quantities of good faith edits.	Low (Dormant)	Mitigate	Unless otherwise requested per wiki, the tool available to those with rollback rights.	Qualitative complaints and anomalies surfaced in data reporting	Implement a maximum number of undos/rollbacks that can be done in a range of time.
Someone is exposed to vandalism or there is a copyright violation in Example Messages provided in-app	A vandal has edited the Example Messages or pasted copyright material in using Community Configuration 2.0 or Translate.wiki, and they automatically appear in everyone's app	Low (Dormant)	Mitigate	Once community configurable, ensure permissions are set for editing with Growth team.	Qualitative complaints	Inrease necessary permissions to edit Example Messages.

How to Follow Along

We have created T322083 as our Phabricator Epic to track this work. We invite you to collaborate with us there or on our Talk Page.

We will provide periodic updates on this page as we make progress. If you are interested in participating in our moderated design feedback sessions, please let us know on our talk page, and we will follow up with you.

We have the following roll out strategy:

2023–2024 time period	Consultation	Released to rollbackers
October 2023	Indonesian Wikipedia
November 2023	Indonesian Wikipedia	Indonesian Wikipedia
January 2024	Spanish Wikipedia, French Wikipedia
February 2024	Spanish Wikipedia, French Wikipedia
March 2024	Chinese Wikipedia, English Wikipedia, Igbo Wikipedia	Spanish Wikipedia, French Wikipedia
April 2024	English Wikipedia, Igbo Wikipedia, Global Sysops and Rollbackers	English Wikipedia, Igbo Wikipedia
May 2024		Available for all wikis

Technical Documentation

If you are interested in seeing how the feature is implemented, please visit the Patroller Tasks Dev Note for more details.

Updates

September 2024

• After conversations with Igbo Wikipedia admins, we changed the availability of Edit Patrol for users on Igbo Wikipedia, and it’s now available to users with 500 edits and 30 days of tenure. T371442

August 2024

• Inspired by this feature, we also added an entry point to the talk page from the Watchlist Diff page. T370697

July 2024 - Results after 30 days

We have results to share 30 days after the release of Edit Patrol to all Wikipedias within the Android App. Overall, we met or almost met 3/5 of our Key Indicators, and 2/2 of our guardrails.

Validation

• KR 1.1: 65% of Target mature audiences that use the tool say they find it helpful for maintaining the quality of wikis and would recommend it to other patrollers
  • Almost met: In an in-app feedback about satisfaction with the feature, 63.6% of users who gave feedback selected Satisfied, 22.7% selected Neutral, 13.6% selected Unsatisfied
  • Qualitatively: an Indonesian Wikipedia editor mentioned they found accessing the talk page templates through the feature helpful, and that it would make them more efficient. They mentioned this feature is the only option for them to check Recent Changes on Indonesian Wikipedia. Another Indonesian Wikipedia editor found the tool helpful for patrolling without needing to carry a laptop. 2 other editors mentioned they are excited to download the app to try the Edit Patrol feature, one mentioned it might make them more active in patrolling.
• KR 1.2: 10% of target mature audiences interacted with another user by sending thanks or a talk page message.
  • Not met: 3.2% of all users sent Thanks, no users sent talk page messages through the feature. We heard from patrollers about the importance of communication along with undo and rollback actions especially for newcomers. However we did not see users sending Talk Page messages along with edits, even though ‘Talk’ was an option in the toolbar. Iterative improvements to motivate users to send talk page messages should be considered for future work.
• KR 1.3: 10% of target mature audiences engage with filter for preferences
  • Met: 23.8% of users engaged with Filters
• KR 1.4.1: 30% of users engage with the tool more than once in a 30 day window
  • Met: 38.1% of users returned on more than 1 day
• KR 1.4.2: 65% of Target mature audiences (potential sysop/rollback Android editors) engage with the tool once in a thirty day window
  • Not met: 30.6% of all potential users engaged with the tool once. This could be due to a few reasons: many users with extended rights already have established workflows on Desktop or Mobile Web, or users with extended rights who have the the Android App are not aware of the feature.

Guardrails

• Guardrail 1: Experienced users without rollback rights, users that have and have not used alternative patrolling tools equally understand the workflow
  • Met: We conducted interviews with 6 Wikimedians who had a variety of experience patrolling with V1 of the feature. Some were admins, and others were experienced editors. 2 interviewees had not used alternative patrolling tools.
    • 6/6 of participants thought that the location of the tool is intuitive.
    • 6/6 could easily understand the 'List of edits' page displayed
    • 6/6 of the participants thought that the information provided through the feature looked fine.
    • 2/6 of the participants didn't understand the title 'ORES score' (they knew good vs bad faith edits but didn't relate that to the ORES score), and we updated the way the scores were presented in response.
• Guardrail 2: We do not receive reports of tool being used to negatively target underrepresented content or contributors based on in-app reporting mechanisms
  • Met: We have not received reports through in app reporting.

Curiosities

• Curiosity 1: How does use of our tool compare to other patrolling tools when looking at MediaWiki Tags (SWViewer, Huggle, and Twinkle)
  • One of our goals in building this tool was that it could work across language wikipedias.
  • In terms of Edits: We saw that 93.1% of Android Patroller edits were made on wikis other than English, comparable to SWViewer where 88.4% of edits were made on language wikipedias other than English. This is markedly different from Huggle and Twinkle, where the majority of edits are made on English Wikipedia

Rollback/Undo Edits by Tool on English vs Other Wikis
Tool	Total Rollback/Undo Edits	% English Wikipedia Edits	% Other Wiki Edits
Huggle	9249	88.6%	11.4%
Twinkle	17133	89.9%	10.1%
SWViewer	3758	11.6%	88.4%
Android App Patrollers	331	6.9%	93.1%

• • In terms of unique Editors, 60% of unique editors on the Android Patrolling tool were editing on a language other than English, comparable to SWViewer where 71% of editors are editing on a language other than English. Huggle and Twinkle’s unique editors/bots were primarily editing English Wikipedia, with only 10% of Huggle’s unique editors editing a language other than English, and 14% of Huggle unique editors.

Unique Rollback/undo Editors by Tool on English vs Other Wikis
Tool	Total Uniques	% English Wikipedia Editors/Bots	% Other Wikipedia Editors/Bots
Huggle	75	89.33%	10.67%
Twinkle	829	85.77%	14.23%
SWViewer	64	28.13%	71.88%
Android App Patrollers (All non bot editors)	23	39.13%	60.87%

• Curiosity 2: Do we see an increase in Undo/Rollback/Thank events
  • We saw an average 7.9% increase in rollback/undo counts compared to previous monthly events. We did not see an increase or change in Thank events.
• Curiosity 3: How popular is this task with our target audience relative to other Suggested Edits task?
  • Edit patrol had 331 edits in a 30 day time period, making up 2% of all suggested edits. Adding image captions has a similar usage rate, at 3.1% of all suggested edits, and 517 edits in a 30 day time period. Edit patrol has a key difference in availability: it is only available to a much smaller audience: editors with rollback rights (other suggested edits are available to users with 50+ edits).

• Curiosity 4: What actions are most popular in the feature?
  • All editors who entered the feature opened an edit from the list of edits, and swiped through edits. 25.4% of all editors completed undo or rollback actions.

Action	% of All Users Completing Action
Swiping through edits	100.0%
Clicking on an Edit	100.0%
Completing an Undo	23.8%
Completing a Rollback	12.7%

• Curiosity 5: How are users interacting with Saved Messages and Example Messages?
  • We have not seen engagement with Saved Messages so far.
• Curiosity 6: How are users interacting with template search and filing out template data?
  • We have not seen engagement with Templates so far.

Lessons Learned

• 30% of our eligible audience engaged with the tool in a 30-day window. When deciding how to release the feature, we listened to community requests to only make it available to users with rollback rights. This resulted in a smaller eligible audience, and we did not have a baseline for frequency of use in a given period. In the future, we should continue to explore how the feature could be made available to more users, especially on small to medium-sized Wikipedias. If community configuration capabilities are added to this tool in the future, it could allow for expanded access: communities could choose to allow editors with a certain edit threshold to have access to this tool.
• We heard from patrollers about the importance of communication along with undo and rollback actions, and we heard requests to add support for Templates. However, we did not see users sending talk page messages along with edits, or using templates within the feature. The feature could have benefitted from an in-app feedback mechanism triggered after a user completed a rollback or undo function, but did not send a message.

June 2024

• We connected with Moderator Tools on the release of the RevertRisk model to RecentChanges, and agreed that we'll wait to confirm model stability and effectiveness with Automoderator on production wikis before assessing the work needed to make the RevertRisk model available via RecentChanges. In addition, we are still monitoring the timing of needing to switch to RevertRisk model as a result of anticipated degradation of the GoodFaith and Damage scores from ORES legacy.

May 2024

• Edit Patrol is now available on production for all Wikis! T350513
• Community ambassador, Bona, shared the feature at the ESEAP 2024 conference, and we heard the following feedback:
  • One Indonesian Wikipedia editor mentioned they found the talk page templates helpful, and it makes them more efficient. They mentioned this feature is the only option for them to check Recent Changes on Indonesian Wikipedia.
  • Another Indonesian editor mentioned it was confusing at first to change their primary language to view the tool (indicating that we could improve the entry point). They found the tool helpful for patrolling without needing to carry a laptop.
  • 2 other editors mentioned they are excited to download the app to try the Edit Patrol feature, one mentioned it might make them more active in patrolling.
• We have preliminary analysis to share after Edit Patrol’s release (acknowledging this was a phased release, we will share numbers 30-days after its release to all wikipedias) T363353
  • Key Indicator 1: 65% of Target mature audiences that use the tool say they find it helpful for maintaining the quality of wikis and would recommend it to other patrollers
    • Actual: 57.1% of users who gave feedback selected Satisfied, 42.9% selected Neutral
  • Key Indicator 2: 10% of target mature audiences interacted with another user by sending thanks or a talk page message.
    • 0% of users sent thanks or talk page messages. We had a discrepancy in our data instrumentation that caused us to over-report the number of messages that were sent through the tool. This was updated to the corrected percent on 10 July 2024.
  • Key Indicator 3: 10% of target mature audiences engage with filter for preferences
    • We saw 1.6% of users engage with Filters
  • Key Indicator 4: 65% of Target mature audiences engage with the tool at least once in a thirty day window
    • We saw 87.5% of those who opened the feature completed successful engagements with the tool.
• We learned we may be able to offer blocking as an action, as requested by some users
• We’ve documented multiple ways that community configuration could benefit this tool in the future.
• Fixed some small issues in Edit Patrol:
  • Make “Next” button active for Saved / Example Messages in Talk flow, even if user has not made change to saved message T362994
  • Navigating back from editing a saved message now returns you to the tab you started on T362996

April 2024

• The team released “V2” of Edit Patrol to target wikis, which includes Saved messages functionality and support for Templates to production, T356774 & T355141.

• Edit patrol is now available on production for Igbo and Chinese Wikipedia, it will be available in production for all Wikis by mid-May.T350512

• We received feedback that we should change “warn” to “talk” to encourage more positive behavior, and we finished updating this throughout the feature: T360449 & T361648

• Outreach:
  • We met with admins from the Igbo community to share the feature and receive feedback on how it lands with a smaller wiki.
  • We shared release information and consultation hours information with English Wikipedia, and on discussion pages of patrolling audiences: Stewards, Global Rollbackers, Global Sysops, Recent Changes Patrol, Small wiki monitoring team, and Coutnervandalism unit. Our consultation hour was held on Friday, April 26.
  • We coordinated with WMFR to share a short async presentation and demo in the French Admins meeting on April 13.

• Learnings:
  • We learned from one English Wikipedia admin that they would value quick access to Blocking users through this tool, and would expect to find it under the user overflow menu. They would also like to see a prompt to quickly add the appropriate talk page message to accompany the block. T363422
  • A global rollbacker shared that they would like to have access to Advanced Reporting and Vetting (ARV), which is supported through Twinkle.
  • We heard another request for the Example Messages provided to be community configurable, highlighting the importance of this work once Community Config supports the apps T358265
  • We heard feedback that reaffirms our choice to make this tool only available to rollbackers at the start, but suggested a threshold to be considered in the future could be 500+ edits and 30 days of tenure. We see another opportunity for Community Config here, supporting wikis in changing permission levels for this feature.T358265

April 2024, Special Mid Month Update

• Edit patrol has been released to all wikis in the Beta version of the app! To view the feature, download or update the Wikipedia Beta app. Open the Edits tab, and look for Edit Patrol under Suggested Edits. The feature is available to users with rollback rights, and will display if you have rollback rights on the language that is set as your Primary in the App.

Edit patrol demonstration on test.wiki

March 2024

• We released the Edit Patrol suggested edit to French and Spanish Wikipedias. T350511
• Development continued for Saved Messages for Edit Patrol (Example messages available in the App) and access to templates.
• While completing development on Saved Messages, we learned that it wasn’t feasible to allow users to edit, delete, and rearrange the prewritten "Example Messages" from within the app in the same way that users can edit their own custom made messages. We updated the designs for Saved Messages to add visual separation between “Your Messages” and “Example Messages” T359209

• We learned we will be able to use the Growth team’s community config to make Example messages available for modification in the app.
• We solicited feedback about how to handle the links to policies within the example messages: we heard from one global admin (a staff member), who recommended against linking to English Policies as the fall-back. We heard from a member of Igbo community that they may choose to keep the links to English policies, and eventually update them. We decided to modify the wording of the English example messages, adding "this wikipedia" in a few places so that the sentences make sense without links. Translation instructions on translatewiki will say, “Please include the relevant language version of the link(s). If no link is found, please do not include a link. Editors may add a link(s) to another language wiki's policy if desired.” We will continue to get more input as we scale to more languages and iterate on how we are treating saved messages for small wikis until community config is available. The example message texts are on a Media.wiki page with links to translations.
• We received feedback that we should change “warn” to “talk” to encourage more positive behavior, and began updating this throughout the feature. T360449
• In our meeting with Spanish Wikipedians, they reaffirmed the choice of Edit Patrol only being available to rollbackers.
• We are sharing the feature to Igbo Wikipedia April 3rd and will have office hours with English Wikipedia April 26nd.

February 2024

• We hosted two synchronous community consultations & recorded a third where we shared the Edit Patrol workflow, and showed designs for Saved messages and templates:
  • French - 21 February 2024 from 16-17 UTC: recording with translation, slides
  • Spanish - 28 February 2024 from 16-17 UTC: English recording, recording with Spanish interpretation(passcode: za2.jeE8).
  • Chinese - shared via asynchronously: recording, slides, transcripts: Simplified, Traditional
• We've started development on Saved Messages T356774 (example messages available in the App) and access to templates T355141, which was well received by French Wikipedia. Draft saved messages are posted here, and we welcome feedback on the discussion page.
• Another community member raised interest in the example warning messages being configurable via the community config.(T358265#9573393), and we are coordinating with the Growth team on this.
• French Wikipedians reaffirmed the choice of Edit Patrol only being available to rollbackers.

January 2024

• Indonesian Wikipedia requested consideration for using Edit Patrol in the upcoming election to fight disinformation and vandalism. Our team had the feature reviewed by our legal team, and volunteers were cleared to use the tool for patrolling during their upcoming election.

• This month we learned that some communities appreciate being able to have speedy deletion as an option in patrolling tools, which we will consider for future iterations of this feature.

• Additionally, our team met with the Growth, Moderator Tools and Machine Learning teams to discuss adding the revert risk model to recent changes filtering. We may be able to switch over from using Good Faith and Damaging ORES scores to the Revert Risk model as we scale to Spanish, French and Chinese Wikipedias in the coming months. This may open the door for an experimental training task for experienced editors without rollback rights.

• We have established demo sessions for our next set of pilot wikis:
  • French Moderators - scheduled for 21 February 2024 from 16-17 UTC
  • Spanish Moderators - scheduled for 28 February 2024 from 16-17 UTC
  • Chinese Moderators - recorded presentation to be shared asynchronous

• We received warning templates from Indonesian Wikipedia, which evaluated alongside other templates to create saved warning messages that will be used in the app.

• Our designer worked on the flow and UI for the saved warning messages and templates, which will be released in the coming months.
  • Saved Messages,T356774

• 
  Choose Talk
• 
  Select message from list of saved messages
• 
  Select from list of example messages
• 
  Create new message
• 
  Review message and make changes
• 
  Save as new message or update saved message
• 
  Review preview and publish (Web view)
• 
  Edit list of messages by swiping left or right
• 
  Reorder or delete multiple messages
• 
  Saved messages are always accessible from the overflow menu

• Template Flow: T355141: The team is working to add support for templates, so patrollers can utilize existing templates in their messages

• 
  Templates will be accessible from the Wikitext editing toolbar
• 
  In Saved Messages (Edit Patrol), create a new message
• 
  Compose message, and open Template icon
• 
  Search for templates by name, recently used are displayed
• 
  Fill out template form and insert
• 
  Message with template inserted. Then follow same steps to, save, preview, publish

December 2023

Patrolling Pain Points on French Wikipedia

We met with a few language communities to get feedback about Edit patrol. The Indonesian Wikipedia community approved the release of the feature for their community and shared which templates are used for warnings, which we will incorporate in a future release. For the month of December we are enabling users to save warning messages on their device (T343841). In January we will begin outreach to Spanish and Chinese Wikipedia to scale to more communities. We’ve already begun outreach to French Wikipedia via the French Wikimedia ED, he shared some pain points identified by French patrollers as it relates to RC Live and SWViewer and other tools as it relates specifically to the French Wikipedia community that we are taking into consideration for Edit Patrol. We are hoping to establish office hours sometime in February with French Wikipedia to discuss Edit patrol.

November 2023

Release to Indonesian Wikipedia

Edit Patrol was released to Indonesian and Test Wikipedia. In order to see the feature in Suggested Edits you must have rollback or sysops rights.

Through this first release, rollbackers are able to:

• Onboarding to the task with an explanation of when to use Thank, Watch, Warn, Undo or Rollback, as well as an explanation of ORES scores. (T343242)

• 
  Onboarding modal 1
• 
  Onboarding modal 2
• 
  Onboarding modal 3
• 
  patrol edit 1
• 
  patrol edit 2

• Compose, save, and publish first warning messages. (T344842)

• 
  Compose new warn message
• 
  Edit new message
• 
  Tapping back
• 
  Empty subject
• 
  Empty Message
• 
  Save changes
• 
  Message published
• 
  Try publishing without message title
• 
  Add message title
• 
  Message saved and published

• Select, edit, save and publish existing warning message (T344842)

• 
  Add message
• 
  Choose message
• 
  Message selected
• 
  Edit message
• 
  Preview message
• 
  Publish, Save
• 
  Save new message Empty Message title
• 
  Preview message
• 
  Save changes message
• 
  Message saved and published

• View information about other editors including their tenure as an editor and their edit count (T343835)

• 
  Menu-list of edits
• 
  User information modal

• Filters for the recent changes feed include language wiki, registration status, experience level, quality score prediction, user intent prediction, automated contributions and significance of edit. The filter page also allows sorting by latest revision. It is also possible to search through edits.

Filter with no results

• An ability to report issues with the filter (T343834)

App bar menu-problem with the feature dialog

We welcome feedback about the feature as we prepare to reach out to additional wikis to scale the feature.

September 2023: Advancing V1 of Patroller Tasks: Unveiling New Features to the Indonesian Wikipedia Community

We continued development on V1 of Patroller tasks and announced the feature to Indonesian Wikipedia. Our engineers demoed the filters for Patroller task V1:

• 
  No results
• 
  Reset filters
• 
  Default filters

August 2023: Project Updates and Technical Considerations

• We shared updates about this project at Wikimania. The recording can be found on Youtube starting at 1:24:30. We received a question about having a central place for Warning messages that can be used across apps. We are planning to work with the Growth team to explore the possibility of centralized warning messages.
• We became aware of the deprecation of ORES and migration to Liftwing which will host the revertrisk model. Due to our team using the MediaWiki API to show the recent edits feed, we will continue to show the Goodfaith and Damaging Models at least in V1. Using the Revert Risk model would require two API calls, which could slow down the feature, this also doesn’t address filtering. If you have feedback about this technical decision, feel free to comment on our discussion page.
• We created designs for Onboarding to the feature T343242.

• 
  Explaining the purpose of the feature
• 
  List of the main actions
• 
  Data collection
• 
  Explaining the 'Quality' score
• 
  Explaining the 'Intent' score
• 
  Indicating where one can submit feedback about the feature

July 2023: Design Research and Interview Results

Usability Testing

A moderated remote usability test was conducted involving six experienced contributors who use Android devices and volunteer their time for monitoring activities on Spanish, French, English, and Chinese language wikis. We selected participants from these wikis due to their high number of editors with sysop/rollback rights, which allowed us to gather usability feedback effectively.

The patrollers provided valuable insights on early designs and the usability of the 'Edit patrol' tool.

During the test, the editors were tasked with completing short assignments using a limited prototype of the tool. They were also asked to provide real-time narration of their experience as they worked through these tasks.

• 
  Find the tool in the app
• 
  Choose an edit from the list of recent edits
• 
  Review the diff and additional information on the page
• 
  Decide whether the edit should be reverted
• 
  Send a warning message to the editor
• 
  Add the page to their Watchlist

For additional details about the usability test, please refer to the following link.

Results

The feedback and valuable suggestions from participants have contributed to enhancing the initial designs. The following improvements were implemented:

• Participants expressed the desire to have tags displayed for each edit on the recent edits page, and the ability to discern if an editor made multiple edits to a single page.

• 
  Previous list of edits
• 
  New list of edits
• 
  Dropdown to see the edits made by the same editor on the same page

• Our goal was to incorporate filters on the page where recent edits are showcased. To determine the filters for the initial version of the 'Edit patrol' tool, we consulted with the patrollers. They suggested including the following filters: languages, user registration and experience, automated contributions, contribution quality predictions, user intent predictions, latest revision, and significance. Subsequently, we designed a page that features these filters, which is presented below.

• 
  Filters

• The diff view page was simplified

• 
  Previous diff view page
• 
  Previous diff view page with full bottom sheet
• 
  New diff view page

• Minor design enhancements have been applied to the process of creating, saving, publishing, and editing warning messages for patrollers. The revised designs are displayed below for your reference.

• 
  Composing a new warning message
• 
  Choosing a saved warning message

• Participants provided further feedback regarding patrolling, which we aim to convey within the tool via onboarding, tooltips, or reminders. The points include:The significance of communicating with empathy to fellow contributors, and the emphasis on prioritising the quality of patrolled edits over their quantity.

The 'Edit patrol' tool is currently in the development phase, and you can track its progress through Phabricator T343224.

February 2023: Initial Design Concepts

The team developed initial design concepts by incorporating insights from existing tools and feedback we gathered on their limitations. We shared these concepts with the Moderator Tools Team and Research Team for their input and feedback.

• 
  Concept for filtering contribution date, ORES Scores or Language
• 
  Design concept for expanding the bottom sheet
• 
  Design concepts for the overflow menu
• 
  Design concept for using templates when leaving a warning

We will use the feedback received to improve the design concepts and establish a protocol for both moderated and asynchronous feedback from app users. Currently, we are initiating direct outreach to schedule moderated feedback sessions and ensure translation mechanisms are in place for those sessions. Once the protocol is available on MediaWiki, we will send out a broader communication to start gathering asynchronous feedback on our talk page.

The design concepts provided are meant to offer examples of how we aim to address the issues faced by other tools. However, please note that we are in the early stages of this process, and the designs will evolve and expand based on the feedback we receive in the coming weeks.

January 2023: Comparative Review and Existing User Research

The team conducted a comparative review to analyze the existing patrolling tools. During the review, we assessed the similarities and differences between the tools and identified which ones were effective on mobile devices.

The complete list of tools we reviewed include:

• Huggle
• Twinkle

• TwinkleMobile
• STiki
• SWViewer
• WikiLoop Double Check
• CheckWiki
• ReWarn
• CheckWiki

Through our evaluation, we found that the majority of tools were optimised for Desktop use, with only three exceptions. Additionally, only two of the above solutions performed well across various language wikis.

Common elements found in these tools were:

• Icons to represent the main action (e.g. revert, speedy deletion, welcome user, etc.)
• Access to templates
• Filters
• List of contributions
• Diffs
• Information about the editor

In addition to the comparative review of existing tools, we also collected recurring requests from patrollers regarding those tools. The common requests include:

• Filter contributions by topics of interest and specific time period
• An easy way to exclude contributions that have already been patrolled
• No requirement for special privileges like rollback rights to access the tools
• An option between "skip" and "revert" for more nuanced actions
• Tools that promote collaboration among patrollers

By taking these requests into account, we aim to develop a comprehensive patrolling solution that addresses the needs and preferences of patrollers across various languages and user experience levels.

December 2022: Proof of Concept Prototype

The team developed a proof of concept prototype to explore the capabilities and possibilities using existing APIs. The prototype allowed us to test and experiment with different functionalities and assess the feasibility of our ideas. This early exploration helps us understand the potential of the tools and refine our approach before moving forward with further development.

• 
  Anti-Vandalism Patrolling task proof of concept
• 
  Anti-vandalism patrolling proof of concept
• Screen recording of prototype

During our development process, we successfully exposed the ORES score, which provides users with an indication of the likelihood that something may be vandalism. This feature allows users to get insights into the quality of edits and helps them make informed decisions during the patrolling process. We would ideally send feedback to the research team's API for improvements, which will be valuable in improving the accuracy and effectiveness of the ORES score.

November 2022: Defining Initial Audiences

The team conducted research (T322065) to assess the number of Wikipedia app users with sysops and rollback rights. The findings revealed that 372 Android app users and 176 iOS users had these rights. Based on this data, we decided to prioritise the rollout of the feature in the Android app, allowing the iOS development team to focus on building the Watchlist feature.

Additionally, we sought to understand the distribution of users with sysops and rollback rights across different languages. This information played a crucial role in determining the initial target language communities for our direct outreach during the initial research phase. By focusing on these specific language communities, we aimed to gain valuable insights and feedback from users who have experience in moderation and patrolling, which will be instrumental in shaping the development of the feature for wider use:

• English Wikipedia
• French Wikipedia
• Chinese Wikipedia
• Spanish Wikipedia

While we are focusing on direct outreach to specific language wiki communities and OS users, we want to emphasise that we welcome and highly encourage feedback and ideas from any app user with patrolling experience. We value the input of Web users as well, but our priority will be to gather feedback from app users.

It is essential to create a unified experience across platforms, and to achieve this, we are in close communication with the Moderator Tools team, who will be building experiences for Web users. By learning from their findings and collaborating closely, we aim to ensure continuity and consistency across both the app and Web platforms. Your feedback and ideas will play a crucial role in shaping the development and improvement of the patrolling feature for all users.