The Signpost

Arbitration report

The Farmbrough amendment request—automation and arbitration enforcement

Richard Farmbrough

Editor's note: the "Arbitration report" invited Richard Farmbrough to comment on his recent request to the arbitration committee. In an effort to represent all sides of the issue, we also asked arbitrators T. Canens and Carcharoth if they would take the time to answer some questions about the case, since they both commented on the initial request. Carcharoth declined, but T. Canens agreed to talk to us from his own perspective.

Richard Farmbrough was set to have his day in court, but as events transpired, this was not to be so. On 25 March 2013, an accusation was made against Farmbrough at Arbitration Enforcement (AE), claiming that he violated the terms of an automated edit restriction. Within hours, Farmbrough had filed his own request with the arbitration committee, citing the newly filed AE request and claiming that the motion was being used "in an absurd way" in the filing of enforcement requests: "I have not made any edits that a sane person would consider automation."

The AE arm of the arbitration committee blocked Farmbrough for one year, after receiving a go-ahead from arbitrator T Canens and without waiting for input from either Farmbrough or the community. The committee, noting that Farmbrough was blocked, then declined to consider Farmbrough's request.

Meet Richard Farmbrough

Richard Farmbrough is something of an icon in the Wikipedia saga. In 2007, Smith Magazine interviewed him as one of the most prolific editors on Wikipedia. In 2011, he was cited by R. Stuart Geiger in "The Lives of Bots" as the creator of the {{nobots}} opt-out template and an advocate of the "bots are better behaved than people" philosophy of bot development. Farmbrough is also credited with coining the word "botophobia", to make the point that bot policy needs to be as responsive to public perceptions as to technical considerations. Farmbrough described himself to the Signpost as "a reader and sometime editor and administrator of the English Wikipedia ... [I've] contributed to and started many articles, worked on policy, edited templates, created and organised categories, participated in discussions, helped new users, run database extraction, created file lists and reports for Wikipedians, done anti-vandal work, and was a host at Tea-house. I also wrote and ran bots."

Genesis

SmackBot: the earliest incarnation of Farmbrough's first bot, Helpful Pixie Bot
Farmbrough's first bot was Smackbot, later renamed Helpful Pixie Bot "to be more welcoming". Helpful Pixie Bot worked mainly on article space, using mostly the AWB (AutoWikiBrowser) program for general clean-up, dating maintenance tags, checking and formatting ISBN numbers, and other tasks that are listed on its user page; it also ran tasks requested by individual editors or projects. Femto Bot was created later, and did more "meta" tasks, such as archiving and maintaining page lists for WikiProjects.

All of the bots' tasks were approved by BAG, the Bot Approvals Group, "although in the less restrictive environment of 2007 a more liberal approach was taken to 'obviously' good extensions of existing tasks than was later the case." Before being submitted to BAG's testing regime, bot tasks underwent a significant amount of manual testing. In one typical case, Farmbrough manually checked and saved more than 3000 edits over the course of six or seven weeks.

None of Farmbrough's bots are currently running. Some of the code and data from his bots is used in other bots, such as AnomieBot and AWB-based bots. AnomieBot has taken over some of Helpful Pixie Bot's dating tasks, but the other general fixes are not being performed.

Dwarves vs gnomes?

So what went wrong? "In September 2010 I made some changes to the general clean-up, there was some opposition and I agreed to revert the changes ... However, an avalanche had been unleashed, and the matter was escalated to ANI. Subsequently I removed all custom general fixes, and rewrote the entire bot in perl, since AWB at that time could not meet the exacting standards that were being demanded. ... One would think that having agreed to do everything asked, and even gone beyond it, the matter would have rested there; but a series of ANI and ARB filings ensued, some rejected out of hand, others gaining traction until by mid-2012 it had become impossible to edit."

As one observer put it, "What we are seeing here is 'The War of the Dwarves and the Gnomes'. Dwarves are editors who work mainly on content, and typically put a lot of thought into each edit; gnomes are editors who work mainly on form, and tend to make large numbers of edits doing things like changing a - to a –. Richard is a Supergnome, and the comparatively small fraction of errors generated by his huge volume of automated edits ended up costing the dwarves who maintain articles an enormous amount of time. Eventually, after repeated failed attempts to rein him in, the outraged dwarves banded together to ban him."

An automation restriction

The outcome of the 2012 Rich Farmbrough arbitration case, along with its subsequent motions, was not at all in his favor. It contained the wording of the automation restriction that has become so controversial: "Rich Farmbrough is indefinitely prohibited from using any automation whatsoever on Wikipedia. For the purposes of this remedy, any edits that reasonably appear to be automated shall be assumed to be so." A later "amendment by motion" stated "Rich Farmbrough is directed ... to make only completely manual edits (i.e. by selecting the [EDIT] button and typing changes into the editing window)".

Is typing four tildes "automation"?
What, exactly, are "automated edits"?
So did Farmbrough break his automation ban? And what exactly are "automated edits"? Opinion was divided over whether automation had been used. Some said there was no compelling reason to believe the edits were likely automated. Others speculated that the edits might have been done with the "search and replace" function in the edit window toolbar, and therefore not prohibited under the restriction. Still others said the edits could be completely manual. (Farmbrough told the Signpost that it was "a manual error incidentally" that gave rise to the AE posting.)

The Arbitration Enforcement administrator, however, stated that "it appears very improbable that this sort of repetitive change was made without some sort of automation, if only the copy/paste or search/replace functions (which are forbidden under the terms of the decision, which prohibits 'any automation whatsoever')", and defined "find and replace" as automation because "it produces the effect of many keystrokes with one or few keystrokes". If "search and replace" is automation, replied the commenters, then so is "copy and paste" or signing posts with four tildes. Farmbrough pointed out that caps-lock also fits the definition of producing the effect of many keystrokes with one keystroke.

Defining automation

What interpretation of "automated edits" is reasonable? We asked Farmbrough if some automated edits are potentially damaging and others not:


Chilling effect on bot operators?

It has been suggested that this will have a chilling effect on other bot operators, that they will be afraid of making mistakes and getting banned. Says one talk page commenter, "A lot of bot ops and potential botops think twice before starting a bot. I have talked with several editors who want too but are afraid if they make mistakes that the zero defect mentality will get them banned."

Arbitrator T. Canens responded:


Does it matter if edits are beneficial?

We did not think to ask whether sub-optimal edits are beneficial, as long as they move the project forward, but both Farmbrough and T. Canens identified this as an issue.

Said T. Canens, "It is very clear to me that the committee in both the initial sanction and the subsequent motion intended to ban all forms of automated editing whatsoever from Rich, regardless of whether any particular automated edit is beneficial. In general, this happens when the Committee determines that 1) the disruption caused by the totality of the automated editing outweighs the benefits of said editing and 2) there is no less restrictive sanction that is both workable and capable of preventing further disruption. In this case, for instance, given the high volume of Rich's automated edits, a remedy that only prohibits him from making problematic edits would be impractical."

Farmbrough stated, "What we should be concerned about is the encyclopedic project, is something someone is doing damaging or benefiting the project? If it is damging we should look at steps to address that, if it is benefiting we should look at ways to improve it further."

Procedural issues about arbitration and enforcement

The Arbitration Enforcement request against Farmbrough was initiated at 10:29, 25 March 2013, and closed less than 13 hours later, at 23:04, with only the accuser and the AE administrator participating. After a request to leave the case open a little bit longer for discussion was declined, discussion continued on Sandstein's and Rich Farmbrough's talk pages.

Farmbrough's block at AE

T. Canens' statement at Farmbrough's Arbcom request that "I think the AE request can proceed as usual", and Richard's subsequent block, received comments at various talk pages ranging from "[it is] somewhat strange that T. Canens should encourage blocking of an editor who has made an appeal to ArbCom" to "the comments from arbitrators seem to say 'block him, we're not going to change the sanction' (T. Canens) and 'we're not going to change the sanction because he's blocked' (Carcharoth and Risker)."

"I was amazed that one arb suggesting Sandstein go ahead was considered authority to do so," Farmbrough told the Signpost. "Even more at the circular argument 'Rich is blocked so the request to remove the provision he was blocked under is moot'".

We asked arbitrator T. Canens why he had Farmbrough blocked while his Arbcom request was still open.


Autonomy of Arbitration Enforcement administrators

There was also some disagreement over the intentions of the arbitration committee with regard to automation and role of AE.

According to one interpretation of the Farmbrough arbitration case, "it isn't the automated editing itself that is harmful/disruptive, and if there is no harm being done here then the 1 year block does not prevent any problems. So in that sense it is neither punitive nor preventative!" and "the Enforcement By block section says 'may be blocked...' which I can't read any other way than to imply that some discretion is given to administrators to not block or to block for a shorter period when, for example, the infraction was so exceedingly minor or when there is no or very little disruption."

According to another view, "the underlying decision of the Arbitration Committee to consider all automated editing of whatever nature by Rich Farmbrough to be harmful, and to ban all such editing. ... Because Arbitration Committee decisions are binding, AE admins in particular have no authority to question the Committee's decisions; they must limit themselves to executing the decisions."

We asked T. Canens if, under these circumstances, "the arbitration committee needs to clarify their intentions about automation and mass editing". Canen replied:


Is there a way forward?

"I just want to get back to editing" says Farmbrough. "Wikipedians do not edit for thanks and barnstars, though they are both nice to receive. It is however a big disincentive to edit, and part of the hostile environment, when there's a constant (and I do mean constant) threat hanging over every editor's head that they're going to have to spend days and weeks fighting off ANI threads and Arbcom cases every time they do something that someone doesn't like."

Given the absence of any other formal mechanism for dealing with automation disputes, that may be exactly what will happen once the block is over.


















Wikipedia:Wikipedia Signpost/2013-06-19/Arbitration_report