User:Risker/History of Oversight on English Wikipedia

In the beginning

The MediaWiki Extension:Oversight was first created and activated in May 2006. It was developed as a temporary hack to allow a few trusted users the ability to remove revisions. It moved the revisions into a parallel table that essentially made the edits invisible to anyone reading the page history, with the ability to manually restore the edits from the database if necessary. As originally published, oversighters could remove the revisions but could not restore them; the restoration was only able to be done by someone with the very highest level of access to the core database. The sole authorized use of the Oversight tool was to remove revisions that contained personal information such as addresses or phone numbers, which at the time was an increasingly common form of trolling. It should also be noted that the community philosophy at the time was much more utopian, and deletion of any material was seen as an option of last resort.

A key weakness of the Oversight tool was that it appeared to misattribute edits in proximity to those that were oversighted. This "true oversight" tool was, to put it mildly, a sledgehammer that created huge holes in the database that could not effectively be repaired, and could only be reversed by the tiny number of developers who had shell admin access and were WMF employees (volunteer shell admin developers did not do it - and even staff developers avoided it like the plague). Because overuse of the tool was perceived to be potentially as harmful as some of the content being oversighted, it was used very conservatively to the point that it was sometimes difficult to get something truly oversighted; the request normally had to come from the person affected, it would often take weeks for a decision to be made, there was only a handful of oversighters, and quite often by the time a decision was made, any potential damage was already done. There is no evidence that anything that was oversighted using the original oversight tool was "un-oversighted" during the period that the tool was in use; it is possible that it happened, but there was no logging mechanism because it would have been done at the shell level.

Another key weakness of the oversight extension was that it could not be used to oversight an entire page; there was always at least one edit still accessible to administrators. In cases where pages were entirely problematic, this often resulted in renaming of pages outside of policy, breaking up page histories, adding spurious edits, and otherwise manipulating content that was often highly problematic.

Administrators from pre-2009 may remember doing a process referred to as "poor man's oversight", which involved splitting the history of the relevant page, deleting the edits with the "problem" edit visible, and then reassembling the "clear" revisions; this was in some ways more effective than the "real" oversight at the time, but also resulted in edits being misattributed, and page histories being contaminated.

The earliest oversighters were almost exclusively members of the Arbitration Committee, although former arbitrators and a few selected bureaucrats also were granted the permission (or allowed to retain it when they left the Arbitration Committee). Many arbitrators never held the tool in the 2006-2008 period. There was a perception that strong technical skills were important in ensuring the tool was properly used, although in reality it was no more technically difficult to oversight an edit than it was to carry out any administrator functions. A requirement put in place shortly after the tool was introduced was that any project had to have a minimum of two oversighters, so that they could act as a check against each other.

Historically speaking, English Wikipedia was the heaviest user of the oversight tool, which had only been used on about 70 Wikimedia projects; at the time that it was superseded by the suppression tools and inactivated, English Wikipedia hosted about 80% of all oversighted material.

The lead-up to suppression

During 2008 and early 2009, a lot of things changed. During the course of 2008, there were increasing calls for deletion/oversight of personal attacks and the like, and also complaints about the response time; the activities of the Arbitration Committee had also become significantly more complex. Therefore, more oversighters were appointed, including oversighters who were not arbitrators. The first call to the community to apply for the permissions took place in July/August 2008. Globally, more projects sought access to the tool, as they too were experiencing increased levels of vandalism and harmful editing that qualified for oversight. The global Oversight policy was expanded to include a wider range of content. As well, there were calls from many communities, including the developer community, to improve the tool so that (a) it was not so destructive, (b) its use was more transparent (even if the edit was not), (c) it could be reversed by a person with the same permission as initiated the event. Somewhat separately but relevant were calls from editing communities to be able to more easily "delete" individual inappropriate but not oversightable edits from articles in a way that did not mess up the page history, retained proper attribution, and was transparent in the page history. Thus, the revision/deletion extension was created, with additional ability for specific users to suppress certain edits. Additional changes to deletion and logging processes needed to be made to allow oversighters to suppress information in logs, to suppress-delete pages, and carry out suppressions in what might be called "oddball" extensions that didn't create accessible log entries or page histories. The new tool, revision/deletion, was unveiled in February 2009; at first, only oversighters, stewards, and WMF staff with authority to carry out "OFFICE" actions had access to it while it was tested and refined.

Technical and practice changes

Almost every arbitrator in the 2009 class actively participated in oversight activities. With the new tool, oversighters were now able to reverse the suppression of edits. And third, there was a rethinking of the nature of what materials should be considered suppressible, and with community consultation and the agreement of the WMF, the types of material that could/should be suppressed was expanded. As well, the page deletion tool was redeveloped to allow full-page suppression; the old oversight tool was not able to do this. The Oversight team continued to grow, and requests for suppression were now accepted from the community at large rather than just the person whose information was "revealed". There was also a new ability to suppress entries in logs, and eventually edit filters and other new software extensions. In other words, both the process and the philosophy behind oversight/suppression changed. As the Oversight team grew and strengthened, and with community support, the practice changed from "oversight only in extremely narrow cases" to "suppress first, discuss within the team and unsuppress if necessary".

In May 2010, the method for requesting oversight/suppression changed. Instead of sending a request to a standard Mailman mailing list, the request was emailed to an OTRS queue to which oversighters were subscribed. This allowed better tracking of requests to ensure that they received a more timely response, and much more strongly encouraged oversighters to review each other's work. The old mailing list was converted to a subscriber-only, non-archiving list where oversighters discuss edge cases, seek review of all oversight blocks, and learn from each other to consider and develop best practices within the team. It is worth mentioning that less than 1% of all suppression actions result in a discussion; the policies and practices have been refined over the years so that there are fewer edge cases, and most new oversighters are already familiar with the policy, and often have personal experience in submitting requests and receiving feedback on them. Oversight/suppression can still be requested directly from an oversighter, usually by direct communication or IRC; however, users are discouraged from emailing an oversighter directly because response time can be considerably delayed. Most requests as of January 2020 are addressed in under an hour, and this has been the case for many years.

Also in May 2010, the English Wikipedia administrators gained access to the revision deletion tools. Between February 2009 and May 2010, the community had built the guidelines for use of this tool by administrators, which included most of the use cases that had previously resulted in the "poor man's oversight" activities in the past, as well as newer uses that reflected changes in both the way that trolls and vandals abused Wikipedia, and in what the community now considered inappropriate material unsuitable for the eyes of the entire internet. Amongst the acceptable administrator uses was material that was suitable for oversight/suppression, as an interim step before an oversighter was able to apply suppression. This helped in reducing the length of time suppressible material was publicly visible.

In 2014, edits that had been oversighted using the original oversight extension were converted to fully suppressed edits, and oversighters gained the ability to manage those edits in the same way as those that had been oversighted using the revdelete/suppress extension. All of those edits were also added to the suppression log. This process was initiated on projects with smaller numbers of oversighted edits in order to more easily perform quality checks; as the project with the most oversighted edits, English Wikipedia was the last project converted. Once this task was completed, the original oversight extension was inactivated. (See a more detailed description of the process here.)

Access to the Oversight user permission

Since 2008, the Arbitration Committee has periodically invited applications from community members to become oversighters. After a preliminary screening by Arbcom, the names of candidates have been presented to the community for further comment. This commentary has been handled in different ways between 2008 and present, but current process is to permit comment and relevant questions from community members.