Wikipedia:Edit filter/RfC

Introduction

In a case involving the use of edit filters, the Arbitration Committee recommended here that "the community is encouraged to establish a policy or guideline for the use of edit filters, and a process by which existing and proposed edit filters may be judged against these." As a result of this, a review of edit filter use has been taking place over the past months which have led to, among other things, the creation of a new noticeboard and mailing list. There was also agreement that a guideline should be formed to reflect current community consensus on how the edit filter should be used.

Editors are invited to vote on whether the 'Edit filter' section below should become a guideline for edit filter use on the English Wikipedia. Feel free to copyedit the proposed guideline, but for clarity do not change it substantially while voting is ongoing; a revised proposal will be made taking into account the opinions raised here if this does not succeed. Note that a number of further discussions have taken place, including the possibility of moving requests for the edit filter manager user right to WP:PERM, which will be addressed after this RfC; this guideline will function as a base from which further proposals could be made.

An archive of the discussions which occurred during the draft phase of this proposal can be found here. Previous discussions on the use of edit filters include the ArbCom case linked above, its talk page, and this idea lab discussion.

Guideline text, now transferred to WP:EF.
The following discussion has been closed. Please do not modify it.
Edit filter The Edit filter is a tool that allows trusted editors in the edit filter manager group to set controls mainly to address common patterns of harmful editing. It automatically compares every edit made to Wikipedia against sets of conditions that are defined in user-created filters. If an edit matches the conditions of a filter, that filter will respond by logging the edit. It may also tag the edit summary, warn the editor, revoke his/her autoconfirmed status, and/or disallow the edit entirely. The AbuseFilter extension was enabled on the English Wikipedia in 2009. The term "edit filter" rather than "abuse filter" is currently used for user-facing elements of the filter as some of the edits it flags are not harmful; the terms are otherwise synonymous. Because even the smallest mistake in editing a filter can disrupt the encyclopedia, only editors who have the required good judgement and technical proficiency are permitted to configure filters. This page does not discuss technical issues concerning the feature; technical information relating to the operation of the edit filter can be found at Extension:AbuseFilter. There are currently 145 edit filter managers. Basics of usage Further information: Wikipedia:Edit filter/Documentation Edit filters are mainly used to identify and mitigate harmful edits by comparing edits with filtering criteria that address patterns of harmful editing. Filters are created and configured by edit filter managers, but they can be requested by any editor. When an edit being saved "triggers" an active filter, the effect depends on a setting associated with that particular filter: Under the strongest setting, the edit is rejected, and the user sees a message to that effect. (A link is provided for reporting false positives.) It is also possible to have a user's autoconfirmed status revoked if a user trips the filter. Under a less severe setting, the user is warned via a customisable message that the edit may be problematic. The user then has the option to either proceed with the save or abandon the edit. Under an even lower setting, the edit is tagged for review by patrollers. Under the lowest setting the edit is merely added to a log. (This setting is also used in tests of new filters.) Recommended uses Except in urgent situations, new edit filters should generally be tested without any actions specified (simply enabled) until a good number of edits have been logged and checked before being implemented in "warn" or "disallow" modes. If the filter is receiving more than a very small percentage of false positives it should usually not be placed in 'disallow' mode. If a filter is designed to catch good faith edits it should not be placed in disallow mode without an appropriate consensus. Edit filter managers should be familiar with alternatives that might be more appropriate in a given situation. For example, problems on a single page might be better served with page protection, and problems with page titles or link spam may find the title blacklist and spam blacklist more effective respectively. Because edit filters check every edit in some way, filters that are tripped only rarely are discouraged. Edit filters should only be set to disallow to prevent edits that substantially all good-faith editors would agree are undesirable, or where a clear consensus has been reached that a specific type of edit should not be allowed. Any doubts regarding setting a filter to disallow should be discussed with other edit filter managers. User right Only members of the edit filter manager group are allowed to modify filters, though all administrators can view private filters. Edit filter managers also have the ability to edit tags. This group is assignable by administrators, who may also assign the right to themselves. The assignment of the edit filter manager user right to non-admins is highly restricted. It should only be requested by and given to highly trusted users; when there is a clear, demonstrated need for it. Demonstrated ability that one can and will use it safely is absolutely critical. This is because widespread disruption of the entire encyclopedia can easily occur—even unintentionally—with the smallest of mistakes in changing edit filters. Therefore, demonstrated knowledge of the extension's syntax and in understanding and crafting regular expressions is absolutely essential. Editors that are not edit filter managers should consider helping out at requested edit filters and troubleshooting at false positives to help demonstrate these skills. Requests for assignment of the group to non-admins can be made at the edit filter noticeboard, where a discussion will be held before a decision is made. In addition a small number of WMF staff have the right, which they may request from the Trust and Safety group, following WMF procedures. If an edit filter manager is misusing the user right, the concern should first be raised with them directly. If discussion does not resolve the issue, a request for discussion or removal of the user right may be made at the edit filter noticeboard. Have a strong password See also: Wikipedia:User account security If you have the edit filter manager user right, please ensure you have a strong password and follow appropriate personal security practices. Because edit filters affect every edit made, a compromised account will be blocked and its privileges removed on grounds of site security. In the unlikely event that your account is compromised, notify an administrator or bureaucrat (for administrators) immediately so they can block your account and remove any sensitive privileges to prevent damage. Requesting edit filters Main page: Wikipedia:Edit filter/Requested Edit filters can be requested at the requests page. Edit filter managers monitor this page and implement edit filters when a good case is made. If there is disagreement WP:CONSENSUS applies. The desirability of an edit filter may also emerge from discussions elsewhere on Wikipedia or through communication on the mailing list. If it would not be desirable to discuss the need for a given edit filter on-wiki, such as where the purpose of the filter is to combat harassment by an abusive banned user who is likely to come across the details of the request, edit filter managers can be emailed directly or on the wikipedia-en-editfilters mailing list at wikipedia-en-editfilterslists.wikimedia.org. If an editor (who need not be an edit filter manager) believes that an existing edit filter is unnecessary, is preventing good edits, or is otherwise problematic, they should raise their concerns on the edit filter noticeboard or directly with the edit filter manager who created or enabled the filter for further discussion. Private filters While filter settings and logs are by default publicly viewable, some are set to be private. For all filters, including those hidden from public view, a brief description of what the rule _targets is displayed in the log, the list of active filters, and in any error messages generated by the filter. Filters should only be hidden where necessary, such as in long-term abuse cases where the _targeted user(s) could review a public filter and use that knowledge to circumvent it. Filters should not generally be named after abusive editors, but rather with a simple description of the type of abuse, provided not too much information is given away. Filter managers may share the contents of private edit filters with non-administrators on the basis of their good judgement. Be careful not to test sensitive parts of private filters in a public test filter (such as 1): use a private test filter (for example 2) if testing is required. Similarly be careful not to post sensitive parts of private filters on talk pages or persistent pages of external sites. Sensitive issues concerning private filters may be raised by emailing filter managers or by contacting them via the wikipedia-en-editfilters mailing list at wikipedia-en-editfilters@lists.wikimedia.org. Mailing list The mailing list wikipedia-en-editfilters is a private list in which only administrators and edit filter managers are subscribers. The list's primary function is for discussion of private filters, both between edit filter managers and with non-admins, who can email the list at wikipedia-en-editfilters@lists.wikimedia.org. The mailing list should not be used as a venue for discussions which could reasonably be held on-wiki. Tools and resources Edit filters sometimes make use of relatively large (though not usually complex) regular expressions (regexes). External tools such as Debuggex can be useful for testing these. Because regexes are extremely fragile and almost any typo in one will cause it to malfunction, use of such a tool is recommended. Notes

RfC

The following discussion is an archived record of a request for comment. Please do not modify it. No further edits should be made to this discussion. A summary of the conclusions reached follows.

First the easy bit: there's clear support for the changes, with the usual caveat that they are a work in progress. Now the tricky points: admins can assign and de-assign the bit at will, and this is an obvious concern as they may not have the technical chops to craft a valid filter. This guideline does not have the force of policy and has no provision for the emergency removal of the flag, which is actually capable of causing more damage at a stroke than almost anything else other than perhaps deleting the main page. That said, I suspect that any amdin who did this would be considered to have "gone rogue" and would be subject to emergency desysopping. Best practice would suggest that no admin enact their own requests without review, obviously. It's not immediately clear how more rigorous change control could be implemented consistent with the current security model of Wikipedia: some filters, by their very nature, must be private and should not be discussed openly if only because of WP:BEANS. In the end, though, all this comes back to the easy bit, which is that this proposal has consensus as a work in progress. It's a start, we're not there yet - perhaps one day this may contain the "must" and "shall" of formally adopted policy. Much thought has gone into this already and doubtless much more thought will go into it in future, and I congratulate the participants on a calm and often insightful discussion. Guy (Help!) 22:17, 15 October 2015 (UTC)

---

Should the above become an official guideline for edit filter use on the English Wikipedia, replacing the information page currently located at Wikipedia:Edit filter? Sam Walton (talk) 17:10, 17 September 2015 (UTC)

Support

1. Support as proposer. Sam Walton (talk) 17:28, 17 September 2015 (UTC)
2. Support. I don't see any guideline as fixed forever, and see any previous finely-tuned changes as being up for amendment, but having a guideline is a good start, and this seems like a reasonable guideline to start with. After so many years it's about time too ... thanks to those who put it together. -- zzuuzz ^(talk) 17:29, 17 September 2015 (UTC)
3. Support. Still needs work (for instance, we need structured testing to insure that edit filters do what they are supposed to do) but a great start, and far better than no guidelines at all. --Guy Macon (talk) 17:59, 17 September 2015 (UTC)
4. Support Full support by me. As an active edit filter manager, I can attest that there is often ambiguity on how to handle certain scenarios: e.g. whether something is worthy of an edit filter, whether actions should be taken against it, or whether it should be private. Most of the time we are able to get by with discussion at WT:FILTER (now WP:EFN) and on IRC, but I find it greatly beneficial to have something down in writing for us and moreover for new edit filter managers to go off of. We have to be reminded that this tool is among the most powerful offered on the wiki, with potential to cause massive (accidental) disruption. Edit filter management should not be a "learn on the job" circumstance. For this reason it seems only logical we should have some established ground rules, and I think in it's current state the proposed guideline does an excellent job of this. — MusikAnimal ^talk 18:31, 17 September 2015 (UTC)
5. Support I'm not sure the current text fully addresses the problem it was meant to solve, but it's a good start. Chris Troutman (talk) 18:33, 17 September 2015 (UTC)
6. Support - It is true, as other supporting editors have noted, that this guideline may need improvement, but it is a reasonable start. Robert McClenon (talk) 18:46, 17 September 2015 (UTC)
7. Support - per all the above. It will certainly be tweaked in future, but it is a very good first foundation to improve transparency and provide clear ground rules. Thanks for putting together this draft, and for edit filter management in general. GermanJoe (talk) 19:37, 17 September 2015 (UTC)
8. Support: It may still need some expansion, but it's a good start. A few more days without any significant opposition or discussion and maybe we can WP:SNOW replace the current page. Esquivalience ^t 00:13, 18 September 2015 (UTC)
9. Support, this is long overdue. Kharkiv07 (T) 00:31, 18 September 2015 (UTC)
10. Support - It can be updated again if necessary. Checkingfax (talk) 00:52, 18 September 2015 (UTC)
11. Support - much needed. Etamni | ✉ | ✓  02:58, 18 September 2015 (UTC)
12. Support It's unfortunate when rules must be used in the place of common sense to stop bad behavior, but too many people believe "You can't prove I'm not allowed to do it because it isn't written down anywhere" to excuse some self-evidently awful behavior. Having written guidelines in place discourages gaming of the system and other bad behavior of the "you didn't say I couldn't so I can" type. --Jayron32 03:14, 18 September 2015 (UTC)
13. Support I don't see any problems here, outside of the issues outlined by davidwr and W.P. Uzer below. But, so far, this looks to be a good start. --IJBall (contribs • talk) 07:06, 18 September 2015 (UTC)
14. Support We may as well have this. Graeme Bartlett (talk) 08:08, 18 September 2015 (UTC)
15. Support It can be helpful in identifying malicious users earlier, and provides data-driven analysis of patterns. This could improve retention of beneficial users. — Preceding unsigned comment added by Cityside189 (talk • contribs) 13:15, 18 September 2015
16. Support - Good start to a guideline that we should have. -- Orduin ^Discuss 19:26, 18 September 2015 (UTC)
17. Support - Long overdue!, Obviously it needs improving but hey it's a start :). –Davey2010^Talk 01:41, 19 September 2015 (UTC)
18. Support - As stated in the introduction, the recent arbitration case demonstrates that there is a genuine need for both a policy/guideline and a process for edit filters. This proposed guideline stands on solid enough principles that I think it's better to have than an information page/essay. However, I'm afraid that it might not be specific enough over what constitutes acceptable usage and what constitutes unacceptable usage. The general goal of reducing disruption/harm to Wikipedia is good for now. Mz7 (talk) 02:12, 19 September 2015 (UTC)
19. Support - I'd like to see some more clarity about patterns of harmful editing - is it spam, vandalism, sockpuppetry, 3RR, support for Trump/Corbyn/political-bogeyman-of-your-choice, what? (To clarify, I'm not after a prescriptive or a proscriptive list, just general guide. It may help forestall frivolous requests.) I would also like to have seen the advice to renounce the bit when it's no longer needed make it into the draft. However, what IS there is sound and makes a good platform to build on. Bazj (talk) 09:00, 19 September 2015 (UTC)
20. Suppport, sounds reasonable to me. Details may be tweaked later if necessary. Huon (talk) 01:08, 20 September 2015 (UTC)
21. Support: this guideline would be a big improvement on the current information page at Wikipedia:Edit filter, and doesn't seem to omit any important information currently on that page. (I presume the text currently on WP:EF would simply disappear if this proposal passed, rather than being moved to Help:Edit filter or elsewhere.) But there is a contradiction between this proposal and WP:EF; this proposal says "Requests for assignment of the group to non-admins can be made at the edit filter noticeboard" while the [old] page currently reads "Requests for assignment of the group to non-admins can be made at Wikipedia talk:Edit filter". Which page should one visit to request the user right? I presume it's the noticeboard but would appreciate it if someone clarified. — Bilorv_(talk)^(c)(e) 10:52, 20 September 2015 (UTC)

    The current information page at WP:EF was really only intended as stopgap; the previous half-info half-documentation page was rewritten to reflect community consensus developed at the idea lab thread with the intention that it should probably be replaced with a proper guideline one day. I wouldn't expect anyone would feel the need to move it anywhere, we can just replace it. And the noticeboard wasn't around when it was written and that page has hardly been updated since this guideline was put together so I've updated it to point to the noticeboard :) Sam Walton (talk) 11:06, 20 September 2015 (UTC)

22. Support - I think this is more a vote to support the idea of a guideline rather than this particular guideline, as surely it will be changed and tweaked just like every other guideline. That said, the edit filter is a possible vector for abuse and damage, and while those incidents are quite rare, a guideline seems only natural as filters are extremely powerful tools. More oversight is also needed, although I don't think it will take extreme measures to achieve this, just as this guideline is not an extreme limitation. Dennis Brown - 2¢ 18:44, 20 September 2015 (UTC)
23. Support - seems like a good place to start placing policies/guidelines for Edit Filter usage. ---- Patar knight - ^chat/_{contributions} 20:51, 20 September 2015 (UTC)
24. Support - A guideline is more official than an information page. Edit fiter access is already governed by a user right request system so it is logical to have it controlled by a more formally recognised text, and it's not as if this would be creating additional bureaucracy. Like any policy or guideline, it can be improved or modified by consencus if that decision is met by a respectable majority at a properly publicised discussion with sufficient particiption to be regarded as a quorum. Kudpung กุดผึ้ง (talk) 00:58, 21 September 2015 (UTC)
25. Support - Support for idea behind guideline, rather than an exact copy of current text. If there are errors or things that need "fixing", then consensus and bold editors will do so. Good start at the very least, Dr Crazy 102 (talk) 06:01, 21 September 2015 (UTC)
26. Support expanding the use of a tool to _target harmful edits with the aim of stopping vandalism and abuse of the encyclopedia; but as others have already noted, proper oversight will be needed. DoctorJoeE ^{review transgressions}/_{talk to me!} 15:31, 21 September 2015 (UTC)
27. Support. This is the sort of thing I was hoping would emerge when drafting the recommendation. Thank you to all who have worked on it. For the avoidance of doubt, this !vote is made in my capacity as an editor and administrator, not as an arbitrator. Thryduulf (talk) 18:16, 22 September 2015 (UTC)
28. Support immediate implementation as guideline: There are some good proposals below for tweaks to this guideline, many of which I would support, but the draft written above is a good starting point and just seems like common sense. I support implementing the text above as the guideline, and any tweaks/improvements can be ironed out on its talk page.--Aervanath (talk) 07:12, 23 September 2015 (UTC)
29. Support This is clearly needed in principle, and this proposal seems like a great starting point. The community can, should, and will continue to develop the guideline as time goes on and the fact that it's not going to be 100% perfectly optimal from the get go is not a reason to oppose. Swarm ♠ 07:29, 23 September 2015 (UTC)
30. Support with the reservation that it will need more work as time goes by. All the best: Rich Farmbrough, 21:19, 15 October 2015 (UTC).
    

Oppose

1. I don't think admins are the right group to manage something like this; they don't necessarily have the required technical knowledge, or the ability to assess whether assignees have the required technical knowledge. It should be run by a fairly small group who really understand the technical issues, similar to the bot approval group. Others would be able to put in requests for filters, which would then be scrutinized, like bot approval requests. W. P. Uzer (talk) 06:11, 18 September 2015 (UTC)

   You do know that's not the change being proposed above, right? The admins right now already have the ability to assign the flag, and have since the Edit Filter was invented... Your opposition makes no sense because if the proposal fails, admins don't lose the right because this proposal doesn't give it to them. This proposal is about adding a list of guidelines for how edit filters are to be used. Please re-read it again. If you do oppose creating a list of rules, please clarify. If you wish to strip admins of the right to assign the EF flag, please start a second RFC to do so, as this one does not deal with the issue you are raising an objection to. --Jayron32 10:27, 18 September 2015 (UTC)

   Well, once something gets enshrined in a "guideline", it becomes understood that whatever is written there has "consensus", and is very difficult to change. Since we're now discussing the introduction of this guideline, it seems a good time to discuss all aspects of it, including whether there is indeed consensus for giving admins this (theoretically enormous?) power. In any case, there aren't really any rules in the proposal, just wishy-washy stuff about what mainly or should perhaps be done, effectively giving admins and other holders of this right carte blanche to do what they like with it. I don't strongly oppose this, but I think it ought to be discussed on its merits rather than just rubber-stamped by a series of people saying "yeah, OK". W. P. Uzer (talk) 10:41, 18 September 2015 (UTC)

   It's not that enormous compared to admins other tools. If you think it is, you're allowed to propose that it be stipped from admins. I'm not telling you you need to just accept it. PLEASE start the discussion. I'm only telling you that this current venue you are raising the point is is the wrong venue to request the change you are making. I'm not telling you to not raise the issue, I'm redirecting you to a more appropriate venue. In other words: If you really want to change this policy, please do so at the correct location. If you want to be completely impotent and not actually change anything, please keep going on about it here, because HERE is the wrong place to get it changed. --Jayron32 15:03, 18 September 2015 (UTC)

   • What Jayron said. I'm for non-admin getting the bit after some vetting, but you still need this guideline to do that. What you need is to work out the logistics, how they are selected (less than an RFA, but more than a request for permissions to one admin). And like he says, do it at the right venue. If it was done right, I'd happily support that proposal. Since Edit Filter is already a dedicated bit, it is very doable. Dennis Brown - 2¢ 18:57, 20 September 2015 (UTC)
   • Per Jayron's You do know that's not the change being proposed above, right?, this oppose is typical and characteristic of the way many serious RfC for system improvements and policy changes get derailed, sometimes to the extent that they just fizzle out and fail by default. I'm sure that a competent closer will know to disregard this !vote. That said, it would appear that W. P. Uzer is not aware of the strict criteria and process under which our admins are (s)elected, part of which assume they (the candidates} have the intelligence, maturity, and above all, our trust, not to use tools that they are not familiar with. Kudpung กุดผึ้ง (talk) 00:44, 21 September 2015 (UTC)

   OK, I see that this has already been decided, anyone who dares to voice even the tiniest criticism of a proposal is insulted and told to take that criticism to somewhere other than where the proposal is being discussed, so that the proposal can get its due rubber-stamp. Which is obviously going to happen anyway regardless of my single entry in the "Oppose" section (why do we even have such sections anyway, when these discussions are not supposed to be votes?). However, it would be more confidence-inspiring if the proposal actually laid down some hard-and-fast rules about what "must" be done to prevent accidental or deliberate misuse of this tool; as it is it just turns a situation in which people can do what they like because there are no rules, into one in which they can do what they like because the rules don't forbid it. If admins were universally super-trustworthy and technically competent like you say, the need for a guideline like this would never have arisen. W. P. Uzer (talk) 06:11, 21 September 2015 (UTC)

   The survey is whether or not the text in its current state is fit for a guideline. Simple as that. Proposed changes are welcomed so that we can reach a consensus on implementing them at a later time. The text touches on what to do if abuse is observed If an edit filter manager is misusing the user right.... I like to think the outlined best practices also do a good job of helping prevent mistakes, but we can't prevent misuse except from not issuing the right to users whom we don't trust, which is implied and is within the scope of the discussion of the editor requesting the right. — MusikAnimal ^talk 15:54, 21 September 2015 (UTC)

2. Simply because I don't think there is enough oversight there. Edit filter managers would effectively be policing themselves as I suspect they'd be the main people active at the edit filter noticeboard. It's not that I distrust the people using the right but rather as a rather specialised and technical group they may have a different view on what is acceptable etc to the wider community. While I accept that that noticeboard is a reasonable first step for requests about the removal of the right and or review of a filter I'd like to see AN as a next step. All admins can see edit filter data, let's make use of that fact. In this regard I think this guideline is worse than the current no guideline / policy situation. In the current situation a user could happily go to AN for further or first review whereas with this guideline in place I suspect any discussion would be quickly shut down as the wrong venue. Hence I see this guideline as introducing less oversight than currently exists so I cannot support it in it's current form. I also think that something like this should probably be policy rather than a guideline given the possible effects of filters (although that would not have made me oppose). Dpmuk (talk) 03:45, 22 September 2015 (UTC)
3. Oppose in its current form, but I would happily switch to support if the issues I raised in the discussion section are addressed. Edit filters are a great tool, but their great power makes control essential. A guideline is needed, but lets get it right before going to press. SpinningSpark 19:22, 22 September 2015 (UTC)
4. Oppose. I agree with nearly all of the points raised by SpinningSpark, and I don't feel articulate enough to add anything new and substantial. —烏Γ ^(kaw), 19:45, 22 September 2015 (UTC)

Discussion

I don't see any mention of emergency removal of privileges, such as for a suspected-compromised account or a this would never happen, I hope editor with this user-right "turning evil". If there is an existing guideline or policy to cover emergency advanced-userrights-removal or if blocking the editor would prevent the use of this user-right, then a simple link to the page describing emergency advanced-user-rights-removal or emergency-blocking will suffice. davidwr/(talk)/(contribs) 05:15, 18 September 2015 (UTC)

This seems fairly straightforward. How to respond to blatant abuse might be implied but other powerful tools make mention of it in their guideline so no harm in adding it here. As for blocking, I went to testwiki to try it out, and just as I suspected only revocation of the admin flag will prevent further self-altered user rights and access to Special:AbuseFilter. Testwiki does not have edit filter manager as a separate user right, but here on enwiki all we'd need to do is remove it for non-admins, otherwise head to #wikimedia-stewards ^connect for emergency desysopping. — MusikAnimal ^talk 05:55, 18 September 2015 (UTC)

• Note I oppose on basic principles snow closing anything that hasn't been open three full days. We don't want to send the message that if you take a two-day break from Wikipedia things will be decided without your input. --Guy Macon (talk) 06:14, 18 September 2015 (UTC)

Edit filters are a dangerous tool that should be handled with care. A filter is essentially a very simple program that can reject human action with little to no oversight; I'm wary of systems who put robots in charge without an overriding mechanism.

If a user finds a false positive in a filter with "disallow" mode, i.e. rule that rejects a valid edit because it triggers an obscure case not predicted by the rule creator, there's very little they can do to solve the problem by themselves. They need to understand the reason why the edit was rejected (which, given the often obscure filter errors, may be a bewildering task), find the filter manager who created the rule, and completely depend on the willingness of this manager to tweak the rule to support the valid case. This breaks the principle of "any one can edit" and having a community of editors collaborating to decide what edits are allowed.

I'm glad that the proposed guideline states that "harmful edits with the aim of stopping vandalism and abuse" should be the primary goal, though what is considered harmful is open to abuse. I'm not concerned with obvious abuse of the system, which would be quickly controlled, but with gray areas where a controversial filter may find substantial, reasonable opposition but get approved anyway; the default position in such cases should be to not enable the automatic system and have humans supervise the situation. For example I've found in the past an administrator in a different Wikipedia who enabled a filter that enforced a specific syntax for some metadata because it was needed by some requisite from Wikidata, and which forbids any edit that doesn't conform to that particular syntax for the metadata. That filter was introduced with little discussion, absolutely no notification to the wide community; and some valid cases which Wikidata didn't support, but which improved the project, were prevented merely because they didn't conform to the Wikidata way of doing things.

I propose the following guidelines before releasing strong edit filters in the wide, to avoid the above problems of placing too much control into the automated filters system:

1. Filters should be primarily _targeted to prevent obvious vandalism of the kinds described as "Silly" and "Spam", which is relatively easy to automate.
2. "Disallow filters" intended to prevent "disruptive edits" should require a formal discussion (a Request for Comments with enough participants) before being enabled, as what may be considered disruptive is ambiguous and subject to opinion. Such discussion should be notified at a general forum like the Village pump, not merely at a forum for edit filter managers.
3. Any filter that gets contested, and was enabled without a discussion of the kind described in step 2, should be immediately disabled until such discussion takes place; not the other way around. That's specially true for filters that prevent the edit from being made, i.e. which can not be overridden by the human.

I think with these safeguard principles in place we could give enough leeway to those editors with permission to create filters, without requiring constant centralized supervision. Diego (talk) 09:21, 18 September 2015 (UTC)

Regarding false positives; all edit filter messages include a link to Wikipedia:Edit filter/False positives where a user can state that that a false positive occurred, and edit filter managers can look into it; no knowledge of the filter or how it operates is required from the user whose edit was flagged. As for requiring consensus for a disallow filter, I can sympathise with the reasoning, but I'm not sure it would benefit the use of filters in practice. I still need to have a think about it though. Sam Walton (talk) 09:37, 18 September 2015 (UTC)

I am more inclined to say that Disallow filters need to be publicized when they are enabled. Not necessarily the technical details which frequently are better off hidden, but at least an announcement in a well-trafficked place, such as the noticeboard or the mailing list, that you have enabled such a filter. I think that allowing for scrutiny is most important here.Jo-Jo Eumerus (talk, contributions) 09:43, 18 September 2015 (UTC)

Such notification is an essential requirement, no doubt. The thing is that it's not enough to prevent problems; the obscure cases that may arise that prevent valid good faith edits can not all be predicted in advance, and will be most likely found by people who didn't know about the existence of the filter when they're in the middle of trying to improve the encyclopedia. They will often be naive editors who know nothing about the filters system, this guideline, nor the policies from which the filter was created. There's great potential for biting the newbies here.

That's why the most aggressive filters should be subject to prior discussion - there are more chances that more people will think of the various possibilities that may trigger a controversial filter in advance; or failing this, the rule to disable an undiscussed filter until discussion takes place will prevent the filter from causing harm if such discussion fails to get a consensus. Diego (talk) 09:55, 18 September 2015 (UTC)

Nothing concerning Wikipedia, including edit filter announcements, should be mailing-list-only, with the sole exception being when a well defined group such as arbcom all agree to use a mailing list and the community supports that group being allowed to have private discussions. There must be a good reason not to use wikipedia pages that anyone can read and with uneditable histories. --Guy Macon (talk) 03:42, 19 September 2015 (UTC)

There is a good reason. I think the guideline makes it pretty clear what it is strictly intended for. Prior to the mailing list we were doing it pretty much the same way by emailing each other individually or on IRC. Some things can't be on-wiki, or it would undermine the whole point of a particular filter being private — MusikAnimal ^talk 04:07, 19 September 2015 (UTC)

Sorry for being unclear. The edit filter mailing list is fine. It's clearly one of the well-defined groups that need to have private discussions. It was the suggestion that edit filter announcements meant for all editor to read be publicized on a mailing list that I was objecting to --Guy Macon (talk) 04:16, 19 September 2015 (UTC)

I think that "announcements" about new EFs would not be very constructive. In terms of impact there's no difference between a new filter an a modification of an existing one. And I don't see what we could say that would be useful without going into detail such as "all edits including the phrase 'burblywurbly' will be blocked"

Even if we said "modern music articles will have more protection against random genre changes" it's not going to stop someone making a non-random change (or we hope not) and if they get a FP we stil need to deal with it. All the best: Rich Farmbrough, 21:18, 15 October 2015 (UTC).

• Just wanted to note that while this proposal's main aim is to put current practices into a guideline, there are one or two things in there that should address some previous complaints about filter usage. The big one is that "If a filter is designed to catch good faith edits it should not be placed in disallow mode without an appropriate consensus" which would, for example, not allow for a filter which disallows editors from adding unsourced awards tables without appropriate consensus for such a filter. I absolutely support further discussion on how this guideline could be refined after (if) it is enacted, but I think simply having these rules as a guideline is an important step, and didn't want the RfC to get bogged down by individual proposals within it. Sam Walton (talk) 13:58, 18 September 2015 (UTC)

  I would have thought that any use of disallow mode ought to require consensus, or at least compulsory review by programming experts, since what it's "designed" to do might very well not be what it actually turns out to do. A badly constructed regular expression might turn out to block editing of Wikipedia entirely. W. P. Uzer (talk) 14:20, 18 September 2015 (UTC)

  That's why we put filters in log-only mode first, to ensure we don't get false positives. It is the responsibility of the edit filter manager to do this, and unfortunately not all filters are meant to be public and hence should not be discussed on-wiki. Also sometimes disruption that's easily preventable via an edit filter comes in rapidly, in high capacity, and we need to take immediate action to put a stop to it; e.g. another major website instructs their user to come to Wikipedia and vandalize using some phrase. I'm not opposed to the idea of making note of which filters do disallow, assuming they are intended to be public, but getting prior discussion for each and everyone may not work out as well as we would hope. — MusikAnimal ^talk 16:04, 18 September 2015 (UTC)

  A good example of this is Special:AbuseFilter/689 where a filter was set up within a day and stopped a lot of vandalism that would have slipped by if we'd had to wait a week or more for a discussion to reach agreement that we needed a filter and that the implementation was good. Sam Walton (talk) 16:35, 18 September 2015 (UTC)

  Another is Special:AbuseFilter/675, which was a result of ClickHole instructing their readers to put "them's the facts!" at the bottom of every article on Wikipedia. I was patrolling recent changes and noticed a flood of this disruption coming through, so I was quick to create a filter for it. This phrasing was very easy to _target with virtually nil false positives, and very high preventive value, so I set it to disallow after only a few minutes of testing. This is obviously a more extreme example, but the point being there are times where we have to be prompt in using this invaluable resource to prevent widespread abuse, and prior discussion is not feasible — MusikAnimal ^talk 18:36, 18 September 2015 (UTC)

The system also allows the revocation of "autoconfirmed" status, but this not in use at the time of writing.- if that were in an article I'd tag it {{when}}. Bazj (talk) 15:23, 18 September 2015 (UTC)

Does anyone know if this feature isn't used because it's not enabled, or because there's no consensus for its use, or simply because we haven't had a need for it? Sam Walton (talk) 15:50, 18 September 2015 (UTC)

I've asked this question before and didn't get any real answers. There is at least one (private) filter where I think this would be useful, but I didn't enable it as it was unclear if it should be used. I found this ancient discussion that might provide some clues — MusikAnimal ^talk 16:15, 18 September 2015 (UTC)

Looking at the Wiki config[1], degroup is not enabled on most wikis including this one. degroup removes all groups in $wgUser->getGroups();. Judging by the source code[2] with getGroups(), "The implicit * and user groups are not included.", which in comparison to other functions suggests degroup will remove autoconfirmed status. I would then say we're not technically allowed to do it, but naturally haven't tried. I do remember an 'incident' involving degroup when the filter was introduced, and there are questions about how to restore the group. Admins (EFMs) can look at the early history of Filter #58 for an example of its use and consequences. With the filter-hit-reporting-bot, I would say it's unnecessary anyway. -- zzuuzz ^(talk) 17:12, 18 September 2015 (UTC)

Also, I am not an expert on these fine details. There is something called blockautopromote, which I can't readily find the details on, nor distinguish from removing the group. It may actually work. -- zzuuzz ^(talk) 18:09, 18 September 2015 (UTC)

It sounds like that might stop the user re-gaining autoconfirmed after it's been removed? I know nothing about this but that's what it sounds like it could be. Sam Walton (talk) 18:20, 18 September 2015 (UTC)

Looking again, I think the blockautopromote action is the only one that's relevant, and I can't see anything that says it's disabled. It's not exactly revocation though. Someone will need to try it on one of their sock accounts to say for sure. I still think it's unnecessary, and we should establish a consensus not to use it except under exceptional circumstances, with consensus on so on, and no false positives. -- zzuuzz ^(talk) 18:57, 18 September 2015 (UTC)

• Change request (or future RfC): Change the Recommended uses section and the corresponding parts of the "Documentation" page to add

  "Except in urgent situations, edit filters must not be set to disallow without thorough testing and a notice to other edit-filter reviewers to give them time to review the filter for technical accuracy. In urgent situations, the notice may be made after-the-fact. Prior to and during the review of an edit filter which is set to "disallow" due to an emergency, the editor placing the edit filter is responsible for seeing that the logs are regularly monitored and false positives are minimized."

  davidwr/(talk)/(contribs) 23:20, 19 September 2015 (UTC)

  This is nicely worded and something I think I agree could be included (in a future RfC). Sam Walton (talk) 11:11, 20 September 2015 (UTC)

  All for this one as well — MusikAnimal ^talk 20:25, 20 September 2015 (UTC)

• Change request (or future RfC): Change the User right section and the corresponding parts of the "Documentation" page to add

  Administrators who do not need this tool are discouraged from assigning this user-right to themselves. Administrators who do assign this user-right to themselves are encouraged (but not required) to put a note on their user page stating whether or not they are available to help in matters where technical competency with the tool is required.

  davidwr/(talk)/(contribs) 23:20, 19 September 2015 (UTC)

  I'm not as enthusiastic about this suggestion, I think we can trust that if administrators assign themselves the userright that they're not going to make any accidental changes, though I could see the benefit for reducing the list of EFMs to only those who have the skill required to work on them. Sam Walton (talk) 11:11, 20 September 2015 (UTC)

  I agree about discouraging adding the right if you're not going to use it. We should somehow make it more obvious that admins do not need the right to see private filters, which I think might be why so many have added it — MusikAnimal ^talk 20:25, 20 September 2015 (UTC)

  This can be part of new-admin-school (does Wikipedia still have that?) and can be put in a mass-message to all existing admins. davidwr/(talk)/(contribs) 21:03, 20 September 2015 (UTC)

  This is a good point. It does still exist, at WP:NAS, and was recently somewhat updated to make sure the instructions still applied and that screenshots were recent. I agree that it should definitely be a part of that; the filter hardly ever gets mentioned as something that's within admin's power to view and use. Sam Walton (talk) 22:04, 20 September 2015 (UTC)

  I've started a page at Wikipedia:New admin/Edit filters. Feel free to help writing it. Sam Walton (talk) 22:08, 20 September 2015 (UTC)

  I'm afraid I don't qualify as I'm not an admin and can't see what admins without the edit-filter userright can and cannot do. If there is a test wiki with identical behavior as en: with respect to the relevant tools that I could be given the admin bit on, I could do comparisons of "with and without the edit filter bit" and use that to help write the documentation. davidwr/(talk)/(contribs) 23:13, 20 September 2015 (UTC)

  No worries, I think I've put most of the relevant info there now so it's now included at WP:NAS. Sam Walton (talk) 10:21, 21 September 2015 (UTC)

  @Samwalton9: The intent was to avoid the issue of editors thinking "oh, he has a technically-oriented bit in addition to his 'admin' bit, so surely he must know how to use that bit," which won't be true if the admin gave himself the bit for the purpose of learning how to use it or to effect an emergency edit filter change that required that bit and only a little knowledge of its use, then didn't bother to remove the edit filter userright when he was done. Discouraging admins from having this bit when they don't need it and encouraging those who have it to indicate their level of expertise will avoid this issue, as well as avoid the issue of admins who have the skills but not the time or desire to help users with edit filter changes being pestered to make them. davidwr/(talk)/(contribs) 21:02, 20 September 2015 (UTC)

• Diego touched on this, but this discussion is not very structured so I'll make my comment here. The guideline should make it explicitly clear what it is permitted to address with edit filters. Yes, I know it says "...mainly used to identify and mitigate harmful edits..." but that is not good enough as "harmful" is too open ended. Anything that breaches any WP guideline can be seen as harmful, but we should not create edit filters that stop users putting periods on the wrong side of brackets, or creating references with the title of the work bolded instead of in italics. As a minimum, the guideline should say something like "deny filters should only be created for situations where the behaviour might in other circumstances result in a block." If it is something that could justify a block, then there is already community consensus for action against it. If we want edit filters for anything else, then either the guideline should explicitly list them so we can discuss them here, or else community consensus should be obtained first on an ad hoc basis. SpinningSpark 19:36, 21 September 2015 (UTC)

  Edit filters should only be set to disallow to prevent edits that substantially all good-faith editors would agree are undesirable, or where a clear consensus has been reached that a specific type of edit should not be allowed. Any doubts regarding setting a filter to disallow should be discussed with other edit filter managers. You make good points though: Noting that such edits would eventually result in a block is something we could add. Also mention that (some, probably most) disallowing filters should be supported by broader consensus and not just edit filter managers, which is something others have brought up — MusikAnimal ^talk 02:08, 22 September 2015 (UTC)

  "edits that substantially all good-faith editors would agree are undesirable" requires an opinion as to what others would agree. That is not good. What all good-faith editors agree to be undesirable is supposed to be set in policy and guidelines, not decided by individuals who can have vastly different opinions and vastly different ideas on what the opinion of others would be. Individual judgement comes in when deciding whether a particular instance is covered by policy and consensus, but there should not be scope for setting new policy or consensus.

  Further, it still does not address my point that not everything undesirable should have an edit filter set on it. Something may be universally accepted as undesirable but an edit filter would still be way inappropriate. Taken literally, for instance, this guideline would permit an edit filter to block use of unapproved date formats. Or the use of title case in sub-headings.

  I also don't like the idea of edit filter managers being the final arbiters of what is acceptable. They are a tiny minority and there is a danger of them becoming a clique with this guideline in place. Decisions should be made with reference to policy. Any referral to EFMs should be on the basis of a request to explain policy, not to create a policy. If a new consensus is needed on an issue, that should be decided at a community forum open to everyone. In this regard EFMs should be acting much like BAG, a filter is only approved if it can be shown that there is consensus for it (it is taken as read that consensus exists if the issue is already in policy or guidelines). SpinningSpark 19:14, 22 September 2015 (UTC)

  Except it wouldn't; "If a filter is designed to catch good faith edits it should not be placed in disallow mode without an appropriate consensus." I understand the general concern though, and I have some ideas for how we could propose an acceptable, stricter, section on disallow filters after this RfC. The primary issue is that we often don't have time to get consensus for every filter and doing so would drastically reduce their usefulness, so wording in a way that respects the community but doesn't stop us responding to things (see above where we give examples of a couple of hastily put together anti-vandal filters) can be difficult. Sam Walton (talk) 19:22, 22 September 2015 (UTC)

  I appreciate the need for fast reactions, but it still needs to be done within the framework of policy. I am not suggesting seeking community consensus or group approval for every new edit filter, only that there must be a policy to cover it. By the way, I also don't think that edit filters should be used in warn mode for good faith MOS issues and the like (logging might be ok). Editors should not be harassed in this way. SpinningSpark 19:31, 22 September 2015 (UTC)

---

The above discussion is preserved as an archive of the debate. Please do not modify it. No further edits should be made to this discussion.