Talk:IP Editing: Privacy Enhancement and Abuse Mitigation/IP Editing Restriction Study/Farsi Wikipedia

Next steps?

Thank you for the well-written report and comprehensive analysis. This work is super important and really well done. The result that account creation still declined even with forcing registration to edit is pretty shocking.

Is there a community vote / discussion underway on Farsi, Portuguese or another other Wikipedia about permanently restricting IP editing? It seems pretty clear this research would strongly discourage any projects from permanently enacting such a policy, but the survey response suggests people's perceptions are different from the real impact on community participation. Steven Walling • talk 18:55, 11 August 2022 (UTC)

@Steven Walling The reason account registration declined there, as opposed to our experience on the Wikipedia in Portuguese, most probably has to do with the lack of any proper directions for IP users to register an account, as we have implemented in our Wikipedia - besides the trend identified on the report, in line with other Wikipedias in the region. If you just block them and do not clearly show how they can participate in the project, the result would always be something like that. In any case, to each one its own reality. IP banning is already permanent at the Wikipedia in Portuguese - and has been like that since 4 October 2020 - and will not be reverted by any action but a decision by the project community, like the one that permbanned IPs. - Darwin ^Ahoy! 01:57, 12 August 2022 (UTC)

Thanks for the insight. Has the trend of account creation increasing continued on pt wiki? Steven Walling • talk 04:15, 12 August 2022 (UTC)

@Steven Walling AFAIK yes, it is still higher than before the ban. In any case, I wonder what's your interest in that metric. Accounts newly created can be, and often are, vandals; spammers; spambots: botfarms , sockpuppets- and all kind of similar useless stuff. One of our initial concerns with the ban was precisely that there would be an increase of account creations due to that stuff, mainly from former vandal IPs and sockpuppets using IPs to bypass the blocks. I have detected myself in the past botfarms using automatically generated accounts that developed inside Wikipedia after the activity of prospective IPs, something that seems to be extinct now after the ban.

In any case, account creation, per se, is meaningless, as it gives no indication if the account is an editor at all, so I'm kind of surprised on your interest on that specific metric. What is interesting to the projects, in general, is not account creation, but the number of active users - either by countering the loss of active users or by increasing them - with the delta being new users and/or "ressurrected" ones, which are both very positive things. On that parameter, you can observe a very striking increase since we permblocked IPs in wiki.pt back in October 2020. But even at the Farsi Wikipedia an increase on active editors can be observed during the months of the IP ban, even if kind of shy, probably because of the factor mentioned above: AFAIK, opposed to wiki.pt, no clear route was given to IP users at wiki.fa to register an account after the block was put in place, and consequently many probably just gave up editing when they saw they were blocked.

Finally, account creation was never an objective of the community decision of banning the IPs. If at all, it was a concern, as mentioned by a number of participants in that debate, who predicted a flood of useless accounts being created, which thankfully has not happened. When taking that decision we had two main objectives:

• Privacy and security of our editors, as IPs are a huge privacy and security hazard for whoever is using them. Unlike registered accounts, which are anonymous by default (at least for everybody until sysop level), IP addresses can easily give away the identity of whoever is using them. And there were already public cases of harassment of wiki.pt editors based on that, including a Twitter account which permanently scanned for certain IP addresses and denounced them as they edited. This is something that was also affecting registered accounts, since it is easy to accidentally log out and not notice it, potentially giving away ones real identity in a second. I swear, I do not understand why, 21 years after that security and privacy hazard was introduced in this project, bringing anything from harassment to physical and live threat over the editors using IPs, the WMF is still allowing it.
• Improve the retention rate of editors, and the overall atmosphere of the project, which was by then, and had been for many years, increasingly swarmed by waves and waves of IP vandalism which were taking a lot of time from established users to counter them, as well as disturbing the work and learning of newbies, to a general feeling of depression and hopelessness related to the project that was leading to many considering abandoning it, and even giving up editing at all. That immediately changed even only 1 day after the ban was implemented.

There were also other secondary objectives, like improving the reputation of Wikipedia - IP editing was generally seen, specially in Brazil, as something bad - even people using IPs vandalized the project to write just that. Anecdotal evidence suggests the reputation of the project has been significantly increasing since then - it was even noticed by the press; eventually increasing the number of new editors (the metric you seem to be interested in, although you seem to be using the wrong indicator) - which happened, as stated; finishing off persistent IP LTAs - and make them more easily traceable by checkusers by forcing them to register an account; and other not so significant stuff mentioned in the debate.

Unfortunately, after taking the decision, we were faced by a blockage from the folk at Phabricator which refused to implement the permblock at Mediawiki level, invoking their own ideological reasons, but that was easily bypassed by the community itself, and, AFAIK, has been working very well since then. - Darwin ^Ahoy! 11:39, 12 August 2022 (UTC)

By the way - it should probably be mentioned that it was never a decision of the Portuguese speaking community that unregistered access to the project would be implemented using such a dangerous and reckless approach as showing to the world ones own IP address. That was forced upon us by the English speaking editors of wiki.en, which discussed that at the wiki.en mailing lists. Since at least 2006 or so there has been talk at wiki.pt VP on rejecting that kind of access to the project, and in 2010 there was even a proposal for banning IPs that never flied for the fear of the kind of colonialist repression by external agents upon our community, as we indeed observed now at Phabricator and other venues. On this subject, I would like to suggest, very clearly, to anyone from outside thinking on interfering on the decision taken by our community: "Stick to your own business." - Darwin ^Ahoy! 12:23, 12 August 2022 (UTC)

• This Restriction does not have consensus in Persian Wikipedia.[1]--Sunfyre (talk) 04:11, 15 August 2022 (UTC)

  Yup, the users voted against banning IP editing again. The biggest reason was that now we have a ML-based revert bot that reverts 3K-4K vandalism every week which is quite a number considering the size of wiki. It's precision is at 83% currently but I'm working on improving it.

  My suggestion to other communities considering this. Try other improvements such as revertbot, blocking abuse filter, RC patrolling, Pending changes, etc. before going with the nuclear option. Amir (talk) 08:46, 15 August 2022 (UTC)

  I'd be more than happy to help any community build a revertbot for their wiki. Amir (talk) 08:53, 15 August 2022 (UTC)

  I assume that "revertbot" here is not mw:Manual:Pywikibot/revertbot.py which basically let you perform what rollbackers do using the command line interface, and that it's a machine-learning (=ML) based one? whym (talk) 11:52, 15 August 2022 (UTC)

  Revertbot? Been there, done that, got the T-Shirt. Tested in different models in wiki.pt, inefficient in all of them, mainly because it basically created a game or challenge for vandals, leading them to vandalize specifically to test or dupe the bot - and making those vandals harder to detect and block by sysops. Also, the idea of a revertbot generally is based on the false notion that vandals stop vandalizing when warned or reverted, which more than often is not the case. - Darwin ^Ahoy! 13:03, 15 August 2022 (UTC)

  @Whym yeah, it's not that pwb one, I wrote a machine learning one from scratch. It took a bit of time (half a week) but definitely worth it. It is much easier to write one (instead of simply using ores which is inaccurate) because now reverts get automatically tagged.

  @DarwIn I remember your case, it had nothing to do with writing ML, you didn't write ML. You hit ores and reverted anything that was scored above 50%. It had lots of issues including the fact that 50% was the wrong threshold and the fact that ores is not very accurate and it becomes less accurate as time goes by. More importantly ores was not for a revert bot. ores was built for filtering recentchanges. If someone spends time and writes one, like Cluebot NG, it works like a charm. Amir (talk) 14:19, 15 August 2022 (UTC)

  Also: Everything we can do with a revertbot can be done, much more efficiently, via abuse filters, which allow for a lot more of control and autonomy, and allow for blocking the vandals before edition and therefore for a more efficient management of vandals, and cleaner history logs on articles. We migrated everything to filters by 2016 or so, and since then all revertbots were deactivated or died naturally for lack of maintenance. It provided for a temporary relief, but was still not enough. Anyway, such patches only helped with one part of the problem, doing absolutely nothing to fix the huge privacy and security hazard which is having one's IP being shown to the world associated with the edition the IP has done. I find this particularly reckless in cases like the Farsi Wikipedia, where someone believing that editing by IP is "anonymous" can potentially lead to the editor being traced by the authorities and killed, if they are based on Iran and the subject of the edition is LGBT+ related. - Darwin ^Ahoy! 13:23, 15 August 2022 (UTC)

  You are referring to a valid problem. As an Iranian gay, I agree on the problem but it has nothing to do with the solution. There is a solution for that and it's called IP masking that WMF hasn't deployed yet for reasons unknown to me.

  Note that I personally added a note to logged-out warning in Persian Wikipedia in 2009 after the protests explicitly mentioning danger of IP editing. Amir (talk) 14:22, 15 August 2022 (UTC)

  @Ladsgroup Yes, IP masking would certainly be a game changer on that. I wonder why IP editing was ever allowed in the Wikimedia projects - Back in 1993 when I was a MOOsaic user, it already had IP masking, assigning random names (stuff like "magenta_guest") to unregistered users, so as to provide anonymity. Personally, I agree that the whole thing of unregistered user vandalism should be rediscussed at wiki.pt after IP masking is implemented, considering the state-of-the-art solutions to minimize the damage it brings to the project. But not until then. As mentioned, editing through IP should have never been allowed in the Wikimedia projects to start with. - Darwin ^Ahoy! 14:39, 15 August 2022 (UTC)

  The technical solution for IP masking is already implemented in the past couple of months. It's just not turned on yet. Amir (talk) 15:00, 15 August 2022 (UTC)

  phab:T300263 Amir (talk) 15:11, 15 August 2022 (UTC)

IPs

First, I like to thanks the authors of the report. I have to mention a few things:

• First, the removal of the IPs had a good result, but now, due to the efforts of the Dear's Amir and Sunfyre there is no need to even close the IPs of the filter breakers, and the robots can handle most of the vandalism and patrolling and guarding. Less is needed
• The second point is that if there is a drop in the participation rate, it is also related to political problems. With the Islamic Republic government's focus on arresting and controlling people in the cyber sector, it takes less time to regularly participate in Farsi Wiki and start systematically changing articles.
• The third point is the sharp decrease and temporary interruption of the Internet itself is an important reason for the decrease in participation.

Personally, due to the existence of the robot to control the editing of unregistered users, I now agree with the complete freedom of IPs. Also, according to the current situation of Farsi speakers in Iran or Afghanistan, I recommend that I have a method to hide or limit public viewing of IPs. Shahnamk (talk) 20:39, 10 February 2023 (UTC)

Talk page behavior

Thank you very much for this research and sharing the outcomes!

I see this was not in the focus of your research, but can you say anything on the usage of talk pages? Did ip-addresses leave comments on the talk pages, about edits they would have done to the articles themselves otherwise? The increase of workload for checkusers is a down side imho, but patrolling and responding to talk pages (often without result because the ip address has already moved on and it can be difficult to engage in conversation) can also mean more work for the Wikipedia's community. Ciell (talk) 20:32, 22 August 2022 (UTC)

@Ciell Article talk pages in wiki.pt are, in general, one of the most failed ways of communicating over there, and almost impossible to track, so I can't tell if they are using it for that - probably not, as most of the people using an IP are using mobile apps and can't see talk pages anyway. We allowed IP access on help request pages and User talk pages, and they do use those regularly. However, in the last days or weeks, a surge of IP attacks and IP caused disruption on help pages is leading to (still kind of informal, AFAIK) discussions on whether we should or not restrict even more IP access to the project. - Darwin ^Ahoy! 11:56, 13 December 2022 (UTC)

Effect on good-faith edits

I'm confused about this highlight:

"However, the restriction also prevented good-faith edits. The total number of content edits was down -24% compared to the previous six months. This was a much larger decline than what we saw in Portuguese. On Portuguese, content edits declined -15% over the unusually high number in the previous six months, but it was generally in line with the previous couple of years. Farsi's decline in edits was well below the previous years."

The analysis on this point makes clear that the other wikis in the region also suffered the same decline. By my understanding, then, the study identified no actual impact on good-faith edits, so it seems misleading to suggest that it did. Emufarmers (talk) 07:43, 2 September 2022 (UTC)

A 24% decline in number of edits of 24% sounds like a big deal, but one has to ask how many of those 24% were vandalistic edits no longer being made? Looking at the absolute numbers rather than percentages is revealing. Total edits declined by around 100k, but reverts declined by about 200k. Since the former figure is much smaller than the latter, that suggests to me that the fall is almost exclusively due to edits we don't want anyway. In fact, since the figures are 100k difference, it suggests that desirable edits have increased by 100k during the test period. SpinningSpark 14:29, 12 September 2022 (UTC)

@Spinningspark: Although the highlight doesn't make it clear, the 24% figure is for "Net non-reverted content edits", so I am willing to accept that there was a decline in the number of desirable (or at least not undesirable) edits. The problem is that the highlight attributes that to the experiment, even though the decline was the same, or worse, on all of the other relevant wikis that were not part of the experiment. Emufarmers (talk) 01:25, 17 September 2022 (UTC)