Wikipedia talk:Requests for comment/Adminbots

Declaring interests
Shouldn't those people running adminbots declare this when they comment here? Carcharoth (talk) 22:43, 7 July 2008 (UTC)
 * I think that they should do, if it shows a possible conflict of interest when commenting.  Al Tally  talk  22:48, 7 July 2008 (UTC)
 * Oh, while we are declaring interests here, I've argued in the past for more openness on these issues. Ironically enough, I became aware of this RfC when I noticed a conversation on MZMcBride's talk page. Carcharoth (talk) 23:03, 7 July 2008 (UTC)

List of adminbots and admins running deletion scripts
I don't think the adminbots I know of are secret secrets, but are more open secrets. Accordingly, I'm listing them here. Any others that people know of could maybe be listed below as well. Carcharoth (talk) 22:47, 7 July 2008 (UTC)

N.B. Some of those listed here may be former or inactive adminbots/script users.

Please add more above as needed. Last updated: 17:56, 10 July 2008 (UTC)


 * Maxim is another off the top of my head. The bot accounts aren't actually admins - you should list the owners, not the bot, because the script is running on the admin account.  Al Tally  talk  22:50, 7 July 2008 (UTC)
 * Could someone work out the best "user" template to use to give links for the logs so people can click through to see the admin actions taken under these accounts? Thanks, Chetblong for adding Nakon. I'm dropping notes off to everyone listed here. I;ll drop one off for Nakon. Could anyone who adds more do the same. Point them here so they can correct inaccuracies and comment. Carcharoth (talk) 23:01, 7 July 2008 (UTC)
 * I think the template would work best here. Thanks for informing the users, I was about to suggest that myself.  Al Tally   talk  23:05, 7 July 2008 (UTC)
 * Is it worth listing the Tor scripts (if that's what they are - only going on vague memories here) and the various developer scripts (mostly now inactive)? I think some other ones got listed at the RedirectCleanUpBot debate. Carcharoth (talk) 23:07, 7 July 2008 (UTC)

Would inactive/former admins Curps and Betacommand be relevant to list here?  Al Tally  talk  23:11, 7 July 2008 (UTC)
 * I suppose so, but if you know for sure they are former/inactive, then say that. Wasn't there a very early blocking bot that dealt with WoW or something? Carcharoth (talk) 23:16, 7 July 2008 (UTC)
 * User:Pending deletion script, User:Portal namespace initialisation script, User:Template namespace initialisation script, are others. -- Chetblong Talk / ARK 23:23, 7 July 2008 (UTC)
 * The initialization scripts aren't really bots. When the portal namespace and template namespace were first created, the MediaWiki software needed to fill in a username in order for logs and page histories to be sensible, and those were the usernames used. --Carnildo (talk) 00:21, 8 July 2008 (UTC)


 * That anti-WoW bot was run by User:Curps, and was informally known as CurpsBot. It also did username blacklisting, and possibly some other functions. --Carnildo (talk) 00:17, 8 July 2008 (UTC)

I think this is a comprehensive list of administrative robots that I've been running so far: Hopefully this can clear the air a little bit, because I don't believe anybody has ever asked me to take inventory in such a detailed manner. east. 718 at 01:15, July 8, 2008
 * RedirectCleanupBot is formally approved for adminship, so isn't really covered by this RFC (imo). – xeno cidic  ( talk ) 23:49, 7 July 2008 (UTC)
 * Deletions: Automatically clears the CSD I4, I5, I6 and I8 backlogs; deletes pages created by spambots; formerly removed orphaned redirect talk pages, and on only one occasion. The imagebot used to be a manual tool, but after discovering that it had an error rate of essentially zero, I decided to run it automatically and make checks on its accuracy after the fact (the false negative rate is astronomical, but that doesn't matter).
 * Undeletion: Mass-restores nonfree images that were deleted under CSD I7 just because of a missing backlink. Other editors are then given the chance to fix these up.
 * Blocking: Puts the brakes on spambots operating from zombie proxies, pagemove vandals, and inappropriate usernames. It usually doesn't get to ever do the latter two though, because the MiszaBot beats it with regularity. Oh, and I also use it to do massblocks, such as cleaning up the aftermath of CheckUser cases involving 200+ confirmed socks.
 * Protection: Migrated the entire English Wikipedia from the old protected titles hack to the new creation protection system. Automatically checks to see if today's featured article is move-protected and will attempt to do so if not; in the event of failure, it will alert admins in their private IRC channel.
 * Edit suppression: Whenever I notice a malfunctioning bot that requires its edits to be reverted, I'll log a bot into my admin account, which will do it and append the "bot" flag to both the rollback and the reverted edit; this will remove the offending bot's edit from recent changes and watchlists. This is also done to the contributions of spambots to existing pages.
 * Thanks for this. It is really helpful to have such information and background. Carcharoth (talk) 08:36, 8 July 2008 (UTC)
 * If anything comes out of this RFC, it's going to be better communication and openness. I have to say I'm really surprised by the way the adminbot community itself has contributed to this RFC, usually it's very hard to get anything out of them, but perhaps we can all understand each other better this time around. -- Chetblong Talk / ARK 12:39, 8 July 2008 (UTC)

East718 above made a comprehensive list of tasks he automates or semi-automates. I run a bunch of twinkle scripts, which do admin tasks at very high speeds, putting me, for practical reasons in the "open-secret adminbot" category". All of the actions I do have to be approved before hand by me; but, afterwards, the script does my job at bot-like speeds. Here are the tasks I've done:
 * Deletions
 * Clearing out I7 backlogs; I made use of User:AzaToth/twinkledelimages.js to clear out the I7 backlog that BetacommandBot created in Februrary. I've used this script for other image categories, too.
 * PROD cleaning: User:AzaToth/twinkledeprod.js. The script automatically gives the reason for PRODing and deletes everything in the category
 * Closed Wikipedia:Miscellany for deletion/User:Grawp with the help of User:AzaToth/twinklebatchdelete.js. I use the script for G8 deletions, as well as other miscellaneous tasks.
 * Protections
 * I use User:AzaToth/twinklebatchprotect.js for adminbot-like protection. There's no specific regular task I do with this script but I've done two major ones. Acalamari asked me to protect all of Grawp's previous userpages and user talk pages after the MfD. The second task is, again, userspace protection, but that is to prevent harassment from trolls. I've used the script for other miscellaneous tasks, too.

Hope this clears stuff up.  Maxim (talk)  16:24, 8 July 2008 (UTC)


 * I dunno if we're including former/inactive admins, but I know that ran a (I think) image deletion adminbot a while back. I don't recall specifics, sorry. —Giggy 08:14, 9 July 2008 (UTC)
 * Going by timestamps, this might be one of the runs (bar the first five, and click "older 50" for more). —Giggy 08:15, 9 July 2008 (UTC)

Any history of significant problems?
Someone would probably have already mentioned it if there were, but has there been any history of problematic adminbot use?-- Kubigula (talk) 04:32, 8 July 2008 (UTC)
 * Yes, several times in wiki history adminbots have either blocked someone or deleted a page incorrectly. And quite a few cases have ended up on ANI, I don't have the time to link to them right now as I'm going off to bed. -- Chetblong Talk / ARK 04:48, 8 July 2008 (UTC)
 * Dunno if I'd call it 'problematic', as it was very easy for me to unblock myself, but, my alternate account has been blocked repeatedly by an adminbot while I was emulating abusive behavior for a non-admin bot :P SQL Query me!  05:53, 8 July 2008 (UTC)
 * The header here says "significant problems". I would say that there have been problems, but whether the significance of the problems should be assessed on a case-by-case basis or overall, is another matter. If people are unaware of the massive scale (and lack of problems) associated with some actions, then they can't fairly make such overall judgments when all they hear about is the problems. This is why openly stating what is done will, in fact, clear the air. I wasn't aware that this RfC was about to be opened, but now that it has, I applaud it. It is also possible that there may have been individual problems that were significant (in the eyes of some), but that shouldn't impact the non-controversial, routine, maintenance or one-off jobs that were and are on problems. But no adminbot or script operator should assume their actions are uncontroversial. The fundamental divide seems to be between those who go ahead with something and then modify things if there are problems, and who also assume that lack of response equals no problems (sometimes a dangerous assumption), and those who would always state their plans first before going ahead with the action. Carcharoth (talk) 08:35, 8 July 2008 (UTC)


 * One that comes to mind is the time someone deleted several thousand images from the precursor of Category:Disputed non-free images without bothering to check if there were any talk page responses or hang-on templates etc. Of those ~8000 or so deletions, perhaps a few hundred were in error (to the extent that the arguments for keeping the image were never reviewed).  Compounding the error was that another bot removed the links to the deleted images from all the pages they had been used on.  It was still a one-off mistake, but a rather large scale one.  Dragons flight (talk) 07:35, 9 July 2008 (UTC)


 * Another example I'd think of is Curpsbot's long-term use of the cryptic "user..." block message when his block bot was issuing username blocks. Dragons flight (talk) 07:35, 9 July 2008 (UTC)

Advertising was light
For this RFC, so I added to Cent and the Village Pump. It was just to a weird collection of user talk pages and the usual admin noticeboards for some reason. rootology ( T ) 13:49, 8 July 2008 (UTC)

What makes an adminbot?
Blatantly stolen from here, but...
 * 1) Admin deletes 1000 pages by manually clicking delete and typing summary each time.
 * 2) Admin makes javascript to have a one-click delete tab, which fills in the summary and deletes when clicked. Admin uses that 1000 times.
 * 3) Admin makes a list of pages, each page has a link to click to delete it. Admin goes through and clicks them all. Now it isn't necessary to load each page in a tab first.
 * 4) Admin makes javascript to automate the clicking, so that the admin makes a single list of pages and clicks once.
 * 5) Same as #4, but instead of javascript the admin uses python. Admin makes a list of 1000 pages and runs a script to delete them from the command line.
 * 6) Admin makes a script that scours the database for pages matching criteria X and deletes them.
 * 7) Admin makes a script that watches recent changes and deletes certain types of pages whenever they are created.

At what point is it an adminbot? I probably use #4 sometimes (via twinkle, hence why I listed myself), but, for the purposes of this discussion, where would everybody define 'adminbot'? SQL Query me! 06:42, 9 July 2008 (UTC)
 * Personally, I should add, my definition probably falls somewhere in the neighborhood of "the process decides on it's own what needs to be done". SQL Query me!  06:44, 9 July 2008 (UTC)
 * If it's making decisions of its volition, it's a robot; cf. numbers 6 and 7 in your example. east. 718 at 07:10, July 9, 2008


 * In my opinion, the "bot-ness" comes in when you stop looking at cases individually and start handling groups of cases "rapidly" through the application of some rule based system to the entire group. That could fall as early as level 3 in your heirarchy if the list of pages is constructed without the admin looking at the pages individually (for example, if he just creates a list out of a deletion category).  I realize that at level 3 there may still be considerable manual input involved (i.e. a lot of clicking), but in my opinion, the process is already "automated" as soon as one constructs an approach that no longer requires human judgment on a case-by-case basis.  Dragons flight (talk) 07:17, 9 July 2008 (UTC)


 * It is possible to apply a regex first, then manually scan the list, then go ahead with a system to automate the boring "click to delete" bits. I assume a regex is the only way a bot can make "decisions". Are there other ways? I would define a bot as a process that doesn't have human oversight. If a human doesn't look at the output or input, but just sets it running, then that is a bot. Halting the process at various points, and having human checks or spot checks, makes it less lacking in oversight. In other words, using programs to apply filters and carry out the actual deletions, but still having a human check every page, is not a bot - that is what I think some of the people using deletion scripts do. Note that in all the cases above, a bot could create a list for human approval. It is only when the numbers get too big and the backlogs too large, or the cases too "simple" or "obvious" that humans tend to get bored or give up, and use or request bots to delete on a regex basis. Carcharoth (talk) 07:20, 9 July 2008 (UTC)
 * So, I think thusfar we're in agreement, an 'adminbot', is one that performs an action without a human making the decision to do so.... I just wanted to make sure. :) SQL Query me!  07:25, 9 July 2008 (UTC)

I believe that the "bot" line is at #6. Twinkle can do #4, and we don't call that a bot. Of course every admin is responsible for actions that they make under their account, whether they are "bot actions" or not. But we can't expect BAG to review things that aren't actually bots. &mdash; Carl (CBM · talk) 03:05, 10 July 2008 (UTC)


 * IMO #6 is where the line is drawn, semi-auto doesn't count as a bot. However there are certain circumstances where 4 and 5 shouldn't be used, but that ends up coming upon the admin's responsibility. -- Chet B. Long Talk / ARK 03:27, 10 July 2008 (UTC)


 * I'm a bit more conservative on this. I'm fine with #2, especially because I use Twinkle to do this myself. *grin* #3/4/5 is where I start to get bothered, because I've run into problems with others doing this. Sometimes when working through the expired prods, I find an article where I can address all the issues, either through changing to a redirect, sourcing, rewriting, etc. I do the work, only to find 5 minutes later that someone deleted it because they had looked at it a while ago (before my changes), added it to their list of mass deletions, and deleted it. I don't think these are good solutions to dealing with backlogs -- if anything, it's chased me away from working on prods, and probably other admins as well.-- Fabrictramp |  talk to me  13:45, 10 July 2008 (UTC)


 * As others, by 6/7 we are clearly into adminbot territory. We could be as early as #1 if the admin is using a bot to identify the pages and doesn't actually review each him/herself.  I have a hard time believing any admin would actually do that with #1 though - if they are going to use a bot to automate selecting the pages and not review them by hand they will go on and let the bot press the buttons.  On the other hand, in 1-5 if the admin actually reviews the pages him/herself and applies judgment before creating the list (with data on version reviewed), and has the bot confirm that the reviewed version is the current version (plus no changes to other pages that might affect the judgment), then I really don't care about the fact that code is using the tool, since it is using it on items where the admin applied judgment and there is no room for change since that would have changed the judgment.  Ultimately, I care about whether human or code judgment is being applied.  If it is code judgment, I want the community to vet the judgment embodied in the code... GRBerry 17:38, 10 July 2008 (UTC)
 * I agree. And I'd like to repeat the point that it is very possible to combine features of bots and humans. Let bots do the filtering and checking and the messy work of generating some stats and lists and collating information. Then let a human look at that list and check a few examples (or all of them if possible), and then let a bot do the grunt work of deletion. Combine the best elements of both - a human's judgment and a bot's reliability. If the system is set up right, you could even (gasp!) let non-admins mark pages as suitable for deletion. Oh, hang on, that's called CSD and PROD tagging! :-) A similar system would work for stuff that's bot delete. Let a bot select the page using filters and whatnot. Let a human confirm the bot's selection. Then let the bot delete. Not all systems will work this way, but a well-designed system can draw on more than just the 1000+ admins, and could use the judgment of patrollers and rollbackers, for example. Carcharoth (talk) 17:45, 10 July 2008 (UTC)
 * IMO, #6 and #7 are adminbots. All the rest are tools used to perform a task quicker, which means that the admin is 100% responsible of any screw-up that might happen (especially with #5, deleting an article that no longer meet a criteria shouldn't be overlooked). -- lucasbfr  talk 10:51, 11 July 2008 (UTC)

Automated deletion of broken redirects
Note: I've split this off, in order for us to continue the discussion of what constitutes an adminbot... Hope no one minds! SQL Query me!  02:43, 10 July 2008 (UTC)
 * I think the distinction about what is and what isn't an adminbot is both important and entirely irrelevant. Bots are capable of doing a number of things that humans may simply skip. To demonstrate this, I'll use an example. About every day, I take a list of broken redirects and run a Python script over it. The script first checks that the page is a redirect; it then checks to ensure that the redirect's target doesn't exist. Once it has done both checks, it does a final check for the last edit made to the page. If the page is older than four days, it deletes the page. If not, the page is skipped. To avoid complications with interwiki redirects, the bot doesn't touch redirects in the User: or User_talk: namespaces. When I used to do broken redirect deletion manually, I would batch open the pages in tabs, and use a one-click JavaScript deletion tab to delete the broken redirect. I would essentially mindlessly click through each tab; if I saw red, I clicked (and deleted), if I saw blue or light blue, I skipped it. The Python script probably pretty clearly would be described as an adminbot. The tabbed deletion probably wouldn't be. But I'm finding it harder and harder to care what the label is and I'm finding it harder and harder to see why it's really relevant. --MZMcBride (talk) 07:33, 9 July 2008 (UTC)


 * Yes, we shouldn't be afraid of adminbots. Their mindless logic can often do a much better job at some tasks than us puny humans.  For example, a bot working on the image deletion backlogs can easily check for recent edits to the image page or the image talk page, while a human is usually too bored to make such checks.  I agree with you that we want more bots.  It would also be nice to have more transparency, and perhaps even a system where more than one person is consulted before deciding what tasks are justifiable.  That being said, if we intend to regulate adminbots at all, we do still need to define "adminbot" in a way that is both useful and unlikely to be ambiguous.  Dragons flight (talk) 07:43, 9 July 2008 (UTC)


 * MZMcBride, did you or the bot check whether the redirects could be fixed to point somewhere else? The assumption you are making is that after four days no-one will bother to fix the redirect. You may also be assuming that it is better to turn such links red, and let people recreate them to the correct place later. Not everyone would agree with you on those assumptions. One thing I would do is also check to see if there are any incoming links to the redirect. The number of redirects also matters here. If it is 50 or a few 100 a day, then that might be manageable by humans. If out of those, a certain percentage were found to be fixable, then you have to weigh up the pros and cons of humans 'wasting' their time on the obvious ones, against the humans fixing the ones they can. It might also be better to find out why such redirects are not being deleted, and to improve the system that way. For example, if they are being deleted because an article was deleted at AfD, then that could be noted by the bot. But say, for example, it was a spelling variant pointing at a sub-article that was spun off another article and then deleted, it requires human judgment whether to have a redirect pointing back to the original article that the sub-article was spun off, or not. The reason it is relevant is that you may be getting such deletions wrong, or missing other ways of dealing with them. Could you provide a list of the redirects you have deleted over the past month using that script? Carcharoth (talk) 08:22, 9 July 2008 (UTC)
 * Perhaps, but my line of thinking is that no admin is currently processing the list in the way you're describing. The four-day check is really to ensure that the page hasn't been vandalized recently and not been caught and reverted. No admin that I know of is going through broken redirects trying to find new targets. Just as no admin is going through unused non-free images and writing rationales for them and finding places where they can be used. Whether the script is deleting or I am, the same judgment is being made. If the redirect is broken, it is deleted under CSD#R1. If it's not broken, it isn't deleted. Perhaps the speedy deletion criteria need more work, but deletions as straightforward as this were what the community itself approved, in its one instance of allowing an adminbot. As for the list of deletions, that should be possible, but it may take a day or two. Or if a Toolserver person is reading this, my user ID is 212624 and the deletion summary is "csd r1" within the past week or two. --MZMcBride (talk) 08:49, 9 July 2008 (UTC)
 * Let's take a closer look at CSD R1: "Redirects to deleted, nonexistent or invalid targets, including redirect loops that do not end with a valid target." - how is the bot judging whether a redlink is non-existent, deleted or invalid? Is the assumption here that a redlinked redirect can only be one of those three cases? What CSD R1 should require is a summary stating why the redirect was a redlink - that would demonstrate that a human has judged the case (a bot could detect that the destination had been moved/deleted, or never existed, or fed into a loop, and record that in the deletion summary - but only humans can judge whether the redirect is invalid or not, for whatever reason - eg. Tree house fun redirecting to Car building timetable). I think what we are running into here is the borderline between CSDs that can be dealt with by bots, and CSDs that still need human judgment. Maybe what is needed is Criteria for speedy deletion using a bot? Oh, and "deletions as straightforward as this were what the community itself approved, in its one instance of allowing an adminbot" - are you really saying that community approval of that very specific set of redirects (different from the redirects you are deleting because they have to have only one edit in the edit history) is a license for you to engage in "task creep" and apply the "permission" to different tasks? I hope you aren't saying that, because that is fundamental to the whole issue here. MZMcBride, would you agree to upgrade your process to detect whether the destination was: (a) moved; (b) deleted; (c) moved and deleted; (d) never existed; (e) fed into a loop; and then ignore the other redirects and let humans judge those - if necessary creating a list for humans to check and validate? Carcharoth (talk) 10:04, 9 July 2008 (UTC)
 * (Adding a late comment in between here:) As I'm partly responsible for the wording of CSD R1, I feel I should note that your interpretation of it is not what I recall it being intended to mean.  "Deleted or nonexistent" is supposed to simply mean "redlink", while "invalid" means titles that MediaWiki won't allow redirects to, such as special pages.  So, yes, as far as I can tell, MZMcBride's deletions are perfectly well within the scope of the criterion.  —Ilmari Karonen (talk) 18:07, 20 July 2008 (UTC)

(unindent) We may want to consider splitting this off into a separate thread as it doesn't really have to do with the definition of an adminbot.

I'll attempt to address each of your points as clearly as I can and I'll also (re-)attempt to put some of my perspective into the discussion. First, if someone creates a redirect like, the script I use would skip it as it would recognize there being an issue with the redirect (namely that it's entirely invalid). Under the CSD#R1 criteria, if someone creates a redirect like, it finds the target redirect and checks the page's existence. I previously forgot to mention that an additional reason for the four-day gap is that, at times (much to my ire), people create 'intentionally' broken redirects. For the most part, they are simply redirects that they will be creating the target for shortly.

The reason I brought up User:RedirectCleanupBot is that, like my script, it is making no judgment about the target. It isn't checking why the target was deleted or whether or not a possible alternative redirect can be made. It is doing exactly what (every) admin does when dealing with broken redirects. It ensures that the redirect is broken, that the page hasn't been recently vandalized, and that there has been sufficient time if someone was creating preemptive redirects for future use.

I've checked hundreds, if not thousands of page histories. Looking at another dimension to this, there may be concerns about deleting only one revision versus deleting six or seven or more revisions. However, as I see it, our role is that of janitors. If a page is created in January, redirected in March, and the target becomes broken in May, it isn't our role to step in and revert back to the March version of the page. That is a content decision for content contributors. Our role is purely janitorial. The redirect is broken, we've written guidelines for this particular type of deletion, the page is deleted.

This discussion, quite frankly, is becoming more and more of red herring in my mind. I fear that any new process for approving adminbots will have the same symptoms that are visible in this discussion. Namely, far higher standards for an automated process than we would ever expect from a human admin. I'll notice that you asked somewhat similar questions in RedirectCleanupBot's RfA, however, you eventually supported. Including after you asked about redirect re-targeting, to which WjBScribe answered that very, very rarely does any admin ever do such a thing. The script that I have been using was written maybe a week or two ago. All other CSD#R1 deletions that I have done have been pretty much manual. I've seen every test case.

One more point I want to mention, and this is directed at Carcharoth, who knows that I have a great deal of respect for him. I took a look at the number of deletions you've done in your time as an admin (longer than mine, I believe) and I thought the tool was broken. In the time you've been an admin, you've deleted 14 pages. While I loathe to turn this into any sort of wiki-dick measuring contest (as I surely don't intend that or want to see that happen), I will say that until you have hands-on experience with these broken redirects (as WjB has, as I have, as east has), the perspective you have on them will be quite different. It's easy, trivial really, to think of a myriad of new methods or detection mechanisms that admins could go through before deleting a broken redirect. However, what I have been trying to underline, and perhaps it's my failing that I haven't yet made the point clear, is that what I'm doing is accepted practice for dealing with broken redirects. It is what every human admin I have ever come across does. And while we can (and perhaps should) expect more from bots and scripts, excessive regulation and requirements will simply lead to people doing what they always do -- ignore the certified process and simply run the script under their main account. --MZMcBride (talk) 20:57, 9 July 2008 (UTC)


 * (we should definitely move this somewhere else) Well, if we are being frank here, I should say that I'd love to write some scripts to do what you do, only better, of course! :-), but I don't have the technical skill to do that. I would also love to try my hand at some manual work in this area, but it is difficult to do this when the bots get there before me. Do you see the point I've just made there? The reason human admins don't do this is because adminbots do it already. I've already said I don't have the technical skill to write a bot or script, so could I ask you whether it is technically possible to get a script or bot examine the history of the page to which a redirect is pointing, and find out whether it ever existed, was deleted, or was moved? I know many human admins wouldn't do this (I would, and I think they should, but then I'm strange like that), and that human admins would get bored with examining 99 redirects before finding the single one that could be fixed (and I do acknowledge your point that it is content creators who should be fixing the redirects, not the bots or the more janitorial of the admins).
 * Taking your comment "a myriad of new methods or detection mechanisms that admins could go through" - the point I was making here is that these detection methods would be dealt with by the bot, all in one go, not by humans. A well-written bot can improve on the job a human admin would do, not just duplicate what a human admin would do. I've recently managed to actually fulfil a commitment I made to start doing some work on wikisource (ie. I do see some things through to a conclusion), so if someone did provide a list of broken redirects, I would (eventually) check them and attempt to get this "hands on experience". You would have to hide them from the bots though, or at least provide me with a list of redlinks after deletion (I don't mind rummaging through the history).
 * Oh, and the experience I do have with redirects is looking for a person under one variant of their name, finding they don't have an article yet, and wishing I could create the variants as redirects to the article that does not yet exist. Many articles about people have multiple redirects pointing at them, and until an article is created, many of the existing links in Wikipedia are different redlinks that should all eventually be pointing at the same article, but which can't be redirected there until the article is created. I recently requested the translation of Amédée Guillemin from the French Wikipedia article, but until that request was fulfilled, I had nowhere to redirect Amedee Guillemin to - what I did was create it as a redirect to Guillemin, and then update that redirect once the translation had come through. Another example was Ramsay H. Traquair - look at the history and the logs.
 * Also consider this situation: you have redirect B pointing at a yet-to-be-created article A. There are 50 incoming links to article A, showing a big demand for an article there. Only 20 incoming links to redirect B, so hey, let's delete redirect B. Then, someone comes along and creates article A. Then later still, someone comes along and sees one of the redlinks for article B (which was a variant name for article A, remember) and creates a new article. My point is that there should be a way to gather together variant redlinks in a way that they all get dealt with by the creation of one article. At the moment, creating redirects is one way to gather the redlinks together (though that misleadingly turns some links blue). Creating an article and creating the redirects is another way. Keeping everything redlinked until an article is created is another. I once tried to create a talk page for the main article (not yet created), and to list the variant names there in the hope that whoever created the page would see the list and create the redirects. But the talk page got deleted as a talk page without a corresponding page. I think the best solution in the end was to list the name and all its variants at "requested articles", in the hope that people seeing one of the links turn blue would then create the redirects.
 * And just to show that this isn't theoretical, have a look at what links here for Eric James Denton, and then at what links here for Eric Denton. There are several pages where people have linked to Eric Denton when they meant Eric James Denton. Sure, the disambiguation page there mentions Eric James Denton, but if there were links to Eric J. Denton somewhere, what should be done in that case? I've found another example as well: Keith Usherwood Ingold (linked from Royal Medal and Davy Medal) and Keith U. Ingold (linked from Henry Marshall Tory Medal) and Keith Ingold (linked from Natural Sciences and Engineering Research Council and Isaak-Walton-Killam Award). Now do you see why people might pre-emptively create redirects?
 * I guess at this point, the most pressing question is where to move the discussion. It certainly shouldn't stay here, though we should try and extract the adminbot relevant bits. Carcharoth (talk) 22:58, 9 July 2008 (UTC)
 * Taking discussion to MZMcBride's talk page. Carcharoth (talk) 10:43, 10 July 2008 (UTC)

Reflecting on possible outcomes
I think it is quite reasonable to assume that this RfC must end in one of the following ways: I have listed them in order of decreasing probability (IMO), but my main point being, all three outcomes have one thing in common: adminbots continue to run (as they should, because they've slowly become an integral part of Wikipedia's operations). The question is: do we want this the easy way or the hard way? Thoughts? Миша 13 22:39, 9 July 2008 (UTC)
 * 1) No consensus reached, status quo retained, adminbots continue to run in a semi-open and unapproved fashion.
 * 2) A pro-bot reform is made, adminbots are reviewed by a competent and knowledgeable body and start to operate in the lights.
 * 3) A more strict policy is adopted, admins start to block others suspecting bot-like activity, adminbots go underground (throttles emulating human activity etc.) but continue to run because there's so many backlogs and too few admins.
 * Maybe you didn't see the notice that This RFC Will Run For 3 Months? :) &mdash; Carl (CBM · talk) 23:53, 9 July 2008 (UTC)
 * No amount of talk will change the fact that one of 3 things can happen: nothing, situation worsens, situation improves. Sorry for playing Captain Obvious here... Миша 13 16:27, 10 July 2008 (UTC)
 * Actually, there are other options, such as "rules change completely" or "all of the above". A change in the playing field is what happened for ProtectionBot, and it was made redundant (I guess you could say the situation improved). But the three options you provide are rather broad. Within "improves" and "worsens" there is a lot of detail that you are glossing over. It is possible that things may just "change", and in that sense things may get both worse and better (eg. some adminbots operators start following a voluntary code of conduct, while others that do borderline stuff refuse to follow the voluntary code of conduct and gradually get sidelined or go underground or rogue, while yet other adminbots are quietly allowed because no-one ever complains about them - a combination of 1, 2 and 3). So I would be wary of offering up too simplistic an analysis. Carcharoth (talk) 17:16, 10 July 2008 (UTC)
 * Yes, I don't know why you would want to close it now when we're getting such great discussion from this, and a lot of adminbot runners are participating. Secondly, there already seems to be a consensus forming for adminbots to be separated from the admin's account, and then made to go through a BRFA. Also I don't think anyone has really requested that they be removed all together, so you're making a rather pointless observation. -- Chet B. Long Talk / ARK 00:13, 10 July 2008 (UTC)
 * No need for you to get emotional about it and treat the RfC like it was your preciousss or something. I'm just pointing out some facts that were obvious to me but apparently are not. Also, I prefer that discussions which have miniscule chance of improving the situation are cut short. Regarding participation, it is actually the adminbot operators that generate the bulk of discussion. Other than that, I see far far less people commenting that one would expect on such a hot subject. Or maybe I was right and nobody cares how the sausage is done as long as it tastes good. Миша 13 16:27, 10 July 2008 (UTC)
 * I thought '3 months' was a joke. There's no un-Godly way this thing can (or should) run that long. In direct response to Chet: a select handful of people have participated in this discussion thusfar. To try to point toward something and say it has consensus would be simply silly and foolish. So, if you're going to call someone's observations pointless, you can start with your own. At the end of the day, any new policy or practice would have be approved by the community, not just those who run adminbots (who are the people largely stating their opinions in this RfC). If possible, I think it would be better if we could get some of the people who have been anti-BAG or anti-bot or anti-whatever (Locke Cole and Monobi come to mind) to participate. Perhaps even contacting some of the people who opposed the RedirectCleanupBot RfA should comment. From what I can see, the views seen here seem rather skewed. As to Misza's comment, I think he's exactly right. There are really only three possibilities going forward, and given the community that we have, I imagine number 3 is far, far more likely to happen than number 2. Though, I personally would have no issue with number 1 continuing; I still don't think a decent argument has been made for change. --MZMcBride (talk) 01:51, 10 July 2008 (UTC)
 * It's changed to 1 month. -- Chet B. Long Talk / ARK 02:32, 10 July 2008 (UTC)
 * That sounds pretty reasonable. It can always be extended. SQL Query me!  03:33, 10 July 2008 (UTC)
 * Regarding #3, I still have some hopes that admins would abide by policy. At least the vast majority of them. Am I too optimistic? --B. Wolterding (talk) 09:05, 10 July 2008 (UTC)
 * Is that optimistic? Without adminbots, the deletion backlogs would be insurmountable, and vandalism would be much more epidemic. When admitbots are all disallowed, only rogues will operate adminbots. – Quadell (talk) (random) 13:57, 10 July 2008 (UTC)
 * Abandon these hopes - when there's a job to be done restraints can (and are) ignored as long as the end result benefits the Projects. Миша 13 16:27, 10 July 2008 (UTC)
 * Just to make that clear, I rather support #2 than banning all bots. But I'm being serious about this point: Whatever the community decides about admin bots, admins are expected to follow it. Doing something against explicit consensus, and even obfuscating it (as #3 suggests), takes WP:IAR at least two levels too far. But perhaps I should rather take that to the main page. --B. Wolterding (talk) 09:12, 11 July 2008 (UTC)
 * The issue at hand is that while we can write as many policies, guidelines, essays, or mandates as we please, the fact remains that it is impossible to detect a human from a bot. Unless we are going to put a CAPTCHA on the delete form, sleep times, number of deletions, time between deletions, etc. can all be forged to look like a human. So, you can certainly expect a high standard of conduct from admins all that you like, but I can guarantee that if a bureaucratic and time-consuming process is put into place to deal with adminbots, quite a few (if not all) adminbot ops will simply do what they currently do. And there's nothing to stop them. --MZMcBride (talk) 00:40, 12 July 2008 (UTC)

Quadell's (former) adminbot
Greetings. I used to run an adminbot: a script under my own account to delete images that had been tagged for deletion. Bless me, father, for I have sinned. Here is my confession.

There used to be a significant backlog for images tagged for deletion. The actual deletion was a tedious and time-consuming process, and there were few admins who (a) knew enough about the details of the IFD process, and (b) were willing to do the grunt work. I did it manually every day, but it took hours. An automated script was clearly needed, so I wrote one.

Now this was the most careful and detailed bot script I have ever written in my life; it's fair to say I was paranoid that it would make a terrible mistake and get me desysoped. I had checks on everything to make sure that if there was any question about the suitability of an image, it would not be deleted. It also included semi-random intelligent wait times, designed to replicate my rate of activity when doing the work manually. It worked like a charm, and the backlog went away.

I wasn't able to discuss this with anyone, since I knew it was against policy. I couldn't get feedback from other users at BRfA, or collaborate with more experienced bot-runners. This meant I had to be all-the-more careful.

After a few months of running this script daily, I found that other admins were doing image deletion much faster and more consistently than before. In fact, by the time my script would run, there would frequently be nothing left for it to do. I strongly suspected that some of these busy admins were using automated scripts as well, but I couldn't know for sure. What distressed me most was that I couldn't really collaborate with them (without risking a reaction of "I'm doing this manually! You're using a bot? I'm telling! I'll have your badge!") There may have been intelligent safeguards that I was using, that other botrunners could have benefited from, and vice versa. It would have made the most sense, in my view, for the task to go through BRfA, with all safeguards published and detailed in public sourcecode, so that we could all compare notes and adopt the safest and best practices. But since the task was unauthorized, that wasn't really possible.

Eventually, other people's bots were willing to run faster and earlier than mine (sometimes even before the due date), so I quit running my script. Hopefully, as a result of this RFC, we can come to a solution that is safer, more efficient, and generally more sane. Thanks for letting me get this off my chest. – Quadell (talk) (random) 13:55, 10 July 2008 (UTC)

P.S. Is anyone else reminded of this clip? ("Adminbots are coming! They're here already!")

Voluntary code of conduct
Does anyone have time to knock up a voluntary code of conduct? Maybe that, combined with "let the adminbot operators regulate themselves within a voluntary code of conduct" might work? I don't have time to write it up myself, but if all the current adminbot operators (and some former ones) signed up to a workable document, it might be a start. Carcharoth (talk) 17:22, 10 July 2008 (UTC)
 * Since no-one took me up on the offer, I went ahead and did this myself. See here. Carcharoth (talk) 21:44, 13 July 2008 (UTC)

MZMcBride's comment
''A lot of great work has been done "secretively." Could the overall situation stand for improvement? Sure. But the community always indicated it didn't want to know what was going on, and so it was left in the dark. --MZMcBride (talk) 16:12, 13 July 2008 (UTC)''
 * MZMcBride, in what discussions on wiki have you seen the community "indicate it didn't want to know what was going on"? Some diffs would be nice. -- Chet B. Long Talk / ARK 02:09, 14 July 2008 (UTC)
 * Adminbots have been running for years. Want diffs? Check any of the contribution logs of any of the editors listed in the pretty table above. I ran one for over five days last week. As though a steady rate of deletions over days is really something a human could achieve? And the community silently watches. I show up in RecentChanges and in the deletion log just like every other admin. It was clear that I wasn't doing the deletions manually. And yet I'm still an admin. I wasn't dragged to AN/I. One user asked me about the deletions, but beyond that, nobody else did. east admitted to running an adminbot on WT:BAG. He actually received explicit approval for doing so. Misza blocks users occasionally. He did so a couple of months ago accidentally. The user was upset and calling for heads, but Misza's still a sysop. east is still a sysop. I'm still a sysop. And this is nothing new – the community has been ignoring the issue for years. Curps used to block users automatically based on their username. Always the same block summary; always done in a robotic fashion. Curps is still a sysop. While "the community" (whatever the hell that's defined as) has never made a single unified diff for me to be able to point to, it's clear that adminbots have and will continue to run with implicit support. --MZMcBride (talk) 02:49, 14 July 2008 (UTC)
 * I think its more that the community knows that arbcom won't do anything, and that wikipedia is completely lacking any rational procedure for removing sysops. But thats just me --T-rex 04:03, 14 July 2008 (UTC)
 * Actually, the community has been very clear on this - namely, I don't think that any version of WP:BOT during the last years contained an exemption for adminbots from the approval process. So the community did want to know, explicitly - it's rather that you didn't want to tell anybody. --B. Wolterding (talk) 11:50, 15 July 2008 (UTC)
 * So basically you are saying that things should stay the way they are just because they have stayed that way in the past? You say that adminbots have run with implicit support, and they will continue to do so, you are evidently ignoring the RFC itself if you think that the secret adminbots are implicitly approved of, considering that nearly 40 people have signed below Majorly's comment stating that all adminbots should be approved through the process which the community will decide. Also the "adminbot" that east718 mentioned on WT:BAG neither blocks or deletes pages, it only move-protects today's featured article which could be done manually just as easily, so it doesn't even require a bot to do the task. And asking about adminbots on the talk page of BAG isn't really a place to find "community" consensus. So what if they've been running for years, vandals have as well. -- Chet B. Long Talk / ARK 18:46, 17 July 2008 (UTC)

(unindent) I'll try to respond to each of the points you made in an orderly fashion so that neither of us gets lost. You said, "So basically you are saying that things should stay the way they are just because they have stayed that way in the past?" That obviously makes little to no sense as what I said previously was "Could the overall situation stand for improvement? Sure." So, we'll move past that comment, as there's really nothing to be addressed.

You said, "You say that adminbots have run with implicit support." Yes, I pretty clearly demonstrated that with my reply above. I strongly urge you to read some of the 40+ supports for Majorly's statement. People have supported a vague notion of approval by an unknown group. One could almost argue that it's simply a re-affirmation of the "radical transparency" mantra that wikis have been following for years. "Secrecy bad, transparency good." Majorly's statement received the level of support that it did because it is two values sentences articulating that transparency is good and a "relevant process" is needed. It's simply shocking that so many people could agree to such language. I think east put it best when he said the notion was "wishful thinking," though others' comments suggested that there's absolutely no consensus for what this "relevant process" would be. Replying to Carcharoth's question, Majorly said that it would probably be a Request for bot approval followed by an RfA. However, had Majorly put that into his original statement, you and I both know that he wouldn't have 40+ supports.

Moving on, you say, "the "adminbot" that east718 mentioned on WT:BAG neither blocks or deletes pages, it only move-protects today's featured article which could be done manually just as easily." Um... duh? Nearly every adminbot task could be done manually 'just as easily.' The point is that nobody wants to or is willing to. And simply because a script doesn't use action=block or action=delete, that automatically means it isn't an adminbot? That's plainly silly.

Your final comment, "So what if they've been running for years, vandals have as well." is from far left field. Whatever point you're attempting to make here has been entirely lost. Adminbots are vandals? Vandals and adminbots have existed for years? I'm a vandal? What? And your assertion that "asking about adminbots on the talk page of BAG isn't really a place to find "community" consensus" assumes that east was looking for community consensus. I think you entirely missed the point I was making by pointing to that thread. --MZMcBride (talk) 19:22, 17 July 2008 (UTC)
 * I must confess that I wasn't looking for "community consensus" when I showed up at BAG's talkpage but rather engaging in self-aggrandizement. That said, some of Chetblong's statement is erroneous: BAG is most certainly aware that my bots delete pages and block users in an automatic, unattended fashion. I'm not sure what you were getting at with the arbitrary limitation that only protecting a single page a day somehow makes the sin lesser, it's still an unassisted robot; in fact, I'd argue that it's worse, since it affects actual, precious articles rather than those spamming proxies and copyright violations we universally loathe. Lastly, I'll ask MZMcBride to look past Chet's disastrously terrible parallel between bot operators and vandals and examine the substance: that the status quo is not necessarily the most prudent one possible. east. 718 at 19:37, July 17, 2008

WP:BOT change
I have removed the "proposed" tag from WP:BOT. Anyone who still has this page on their watchlist may want to comment at WT:BOT. &mdash; Carl (CBM · talk) 01:25, 17 September 2008 (UTC)

Closed
I have closed the RFC, something that I should have done in August. I would appreciate if all further discussion would move to WT:BOT where there is already discussion about the proposed policy. --Coffee //  have a cup  //  ark  // 13:26, 25 September 2008 (UTC)