Wikipedia talk:Bots/Cosmetic Bot Day

To do

• Finalize this page in general.
• Decide if we want to decide how many each bot get on COSDAY individually, set global limits (everyone bot gets the same limit), or have an unrestricted number of edits for the entire day (subject to editing rate limits).
• WP:BOTNEWS call for BOTREQs/COSDAY bots to be made.
• Decide on when COSDAY should be held. — Preceding unsigned comment added by Headbomb (talk • contribs) 23:09, 7 December 2020 (UTC)

Basic coordination

I made this page to have a central place to organize things for a COSDAY trial, assuming there's consensus for a such trial. Feedback welcome. Headbomb {t · c · p · b} 20:16, 7 December 2020 (UTC)

Time slots

The initial proposal of a single calendar cosbot day presents a number of problems. It may cause watchlist overload from too many bots running at the same time. Also, it is unfriendly to bot ops who are forced to work at a certain time/date they may be unavailable. An easy alternative is after a cosbot is approved at BRFA, the botop adds to a schedule (this page) giving the bot name and date they plan to run, preferably not overlapping with any other cosbots. And that's it. Each month the botop wants to run again, they re-add to the schedule but must be > X days from the last time it ran (whatever the wait period is, like 30 days). The schedule serves multiple purposes as informing the community what is being done, and allows BAG to monitor who is doing what when. If a cosbot is found operating outside their posted schedule it can be revoked. -- GreenC 05:46, 8 December 2020 (UTC)

I think the place to make this argument is at the RFC, not here. The RFC clearly proposes a single (periodic) day for all cosmetic bot day activity. – Jonesey95 (talk) 05:49, 8 December 2020 (UTC)

The RfC has (re)closed now. It says "preference expressed for bots/cosmetic bot day structures which will minimize watchlist disruption". -- GreenC 06:12, 8 December 2020 (UTC)

While my personal preference would be to spread things over multiple day, this wasn't really explored in the RFC. My gut feeling here is to have a trial (it's not particularly hard to have a bot/process automatically start at a specific time), and then we'll see if the community wants to spread things out more or keep them concentrated on a single day (monthly, quarterly, semi-annually, etc...). For minimizing watchlist disruption, I feel that's what editing limits are for. We could have a simple hard limit (e.g. 1,000 per COSDAY per bot). We could have a random sort in the order a certain bot edits to avoid article clusters. We could have a per-bot custom limit (bot A can do 100 edits, bot B can do 1,000 edits, bot C can do 5,000 edits). We can mandate that multiple bots are merged [e.g. a WP:CHECKWIKI bot covering all tasks] to maximize the per-edit value, as well as add WP:GENFIXES to make fewer cosmetic edits. Headbomb {t · c · p · b} 06:50, 8 December 2020 (UTC)

Personally, I'd have a 1,000 edit per bot trial limit by default for COSDAY, and then customize the limit on a per-bot basis when there's a need. I.e. if the total task would cover 1,200 articles, might as well do the extra 200 articles. If the total task could cover 300,000 edits, then stick to the 1,000 per day limit for the trial. Headbomb {t · c · p · b} 06:52, 8 December 2020 (UTC)

Keep in mind, these bots have community approval, the community wants these bots to run. -- GreenC 17:11, 8 December 2020 (UTC)

That very much depends on the bot. Right now, no bots has been proposed let alone approved for COSDAY. We're also trialling COSDAY, and wanting to do so in a way that doesn't cause major disruption. Headbomb {t · c · p · b} 17:20, 8 December 2020 (UTC)

If a bot is approved at WP:BRFA the community wants it to run. There is no more watchlist disruption than any other approved bot. Tight regulations and needless complications will dampen if not eliminate bot authors from putting the time and effort in. The community feared a stamped of bulls (typical FUD), so far it is crickets. The restrictions are in-built to a bot "day". For example Monkbot has been running for weeks now and I see maybe half a dozen edits a day on my watchlist. The complaints with Monkbot are not because of 1 day but the day after day. A half dozen watchlist notices on one day is not going to trigger community outrage. We need to encourage and entice participation not limit it based on what might happen, unless a reason comes up in practice. -- GreenC 18:28, 22 December 2020 (UTC)

The issue with a single day is that operating at any feasible rate still doesn’t do enough articles. Really this depends on the bot, as to how much watchlist spam it could do. The smarter idea is along Monk, using randomisation for large scale bots and rate limits for smaller ones. Much better for WL spam + protecting against vandalism. That’s probably the route to explore making permanent post trial. ProcrastinatingReader (talk) 10:25, 8 December 2020 (UTC)

"Enough articles", well it's a trial. Just because something can be done 300,000 times doesn't mean it should be. It's got to be weighed against urgency (low) and usefulness (depends on the task). Randomization is good for when you've got edits to make on 10,000+ unrelated articles. If it concerns a specific infobox, randomization won't do much to avoid targeting clusters of related articles since they'll be related by topic already. There's also a huge 'backlog' of 20 years of things to address, and it's probably best to stagger these updates over multiple runs to not have something like COSDAY 1 = 2,000,000 edits across 30 bots, and COSDAY 2 = 75,000 edits for the same 30 bots. Headbomb {t · c · p · b} 15:29, 8 December 2020 (UTC)

If it concerns a specific infobox, randomization won't do much agree, but that's why I say rate limits for smaller ones. Niche tasks will probably have to be rate limited on CBD as well, to avoid totally flooding watchlists (although this may be unavoidable for such tasks). Otherwise it doesn't matter how many CBDs one does, a reasonable percent of eligible articles will never be edited. Just because something can be done 300,000 times doesn't mean it should be. Not on this trial (that would be almost 4 edits per second), but over time, yes? I can't really think of any reason why X eligible article should get an edit but Y eligible article shouldn't?

None of this is directly relevant to this trial though, but I think post-trial we should re-evaluate these aspects. ProcrastinatingReader (talk) 18:43, 8 December 2020 (UTC)

One edit per article

During the RfC, some users mentioned the idea of having at most one edit per article performed during the bot day. The problem would be to coordinate all the bots. Just a thought: what if a bot checks that the timestamp of the last revision is older than the bot day beginning? It will of course prevent some edits that could have been done otherwise (but they could be done in a later bot day). --NicoV ^{(Talk on frwiki)} 14:19, 8 December 2020 (UTC)

Not viable for a great deal many bots, and also an unnecessary limitation. Headbomb {t · c · p · b} 15:25, 8 December 2020 (UTC)

How is it not viable? I'm pretty sure that AnomieBot waits for a period of time before adding date stamps to templates. – Jonesey95 (talk) 15:47, 8 December 2020 (UTC)

As an example, any WP:AWB-based bot cannot check timestamps, or edit histories, or anything like it. Headbomb {t · c · p · b} 17:17, 8 December 2020 (UTC)

@Headbomb Are you sure this is not possible with a module? 𝟙𝟤𝟯𝟺𝐪𝑤𝒆𝓇𝟷𝟮𝟥𝟜𝓺𝔴𝕖𝖗𝟰 (𝗍𝗮𝘭𝙠) 15:50, 13 March 2021 (UTC)

That suggestion made no sense. If a bot has community consensus (ie. "we the community want this bot to run") there is no reason to kneecap it, unless that is part of the consensus discussion for the bot. If an article requires 10 changes you make 1 edit with 10 changes, not 10 edits of 1 change each. And 10 edits is low, my bot sometimes makes edits that are over 100 changes. It would make it nearly impossible to complete the approved bot task, to go back and reedit the same article every month it could take years to complete meanwhile irritating everyone with watchlist churn and history diff bloat. And no bot operator would bother running a bot under that arbitrary limitation. -- GreenC 16:47, 8 December 2020 (UTC)

You misinterpreted my post GreenC. Of course, grouping modifications in one edit is better, but we won't be able to group modifications from different bots in the same edit, so my suggestion was for articles where several bots would need to edit... --NicoV ^{(Talk on frwiki)} 17:04, 8 December 2020 (UTC)

There's no real reason to limit multiple bots editing the same page in one day. Especially during a trial where we want to see if this is an actual issue. Multiple bots already edit the same page in one day all the time, so I don't really see why it would be an issue here. The reason AnomieBot waits a little bit (15 or so) is because it's triggered automatically when placing undated tags like {{cn}}, and waiting reduces edit conflicts because when someone places a {{cn}} tag, they often do a variety of cleanup afterwards. Headbomb {t · c · p · b} 17:12, 8 December 2020 (UTC)

I'm ok with not having this restriction, I posted this idea just because some users voiced their concern seeing the same article edited by different bots during the bot day (with the impossible suggestion to group the modifications into one edit even if they were coming from different bots...). I also think it's easy to do for some bots (at least for WPCleaner, it won't take me much time to add this option). --NicoV ^{(Talk on frwiki)} 17:35, 8 December 2020 (UTC)

Can an AWB-based bot definitely not do this? I mean, modules allow for the writing of arbitrary C# code, right? And there's a Skip parameter which is a boolean. So, someone can write a module that does the check + skip and everyone else can use that same module? If we're going with a rate limit of 1,000/day for the trial I think this is a good idea, since such a skip isn't a meaningful disturbance to the bot run. It can be preferred but not required. It'd be more easily implemented for bots using a popular programming language (eg Python) where it only takes one person to write the logic for this, and the rest can paste it in. ProcrastinatingReader (talk) 18:46, 8 December 2020 (UTC)

Maybe there's a way of doing so with AWB, but it's certainly beyond my own skills to do so, and adds unnecessary complexity to the trial. Having two or three bot edits in one day on a handful of pages is not something that needs to be defensively coded against. WP:KISS applies. Headbomb {t · c · p · b} 20:08, 8 December 2020 (UTC)

Well, as I say, only one person has to write the code to do it. Then everyone else can just copy+paste it into their "Make Module" thingy and add in a call at the top of ProcessArticle along the lines of Skip = !CBDShouldProcessArticle(signature). I think if someone wants to write this (perhaps for their own bot), they can make it available for everyone else, and we could then mandate it for all AWB users just because it's so simple to paste in. That's assuming we actually agree with the principle of 1 bot edit per article. I can see the arguments for/against. ProcrastinatingReader (talk) 20:43, 8 December 2020 (UTC)

Recommend wait until there is an evident problem of clashing cosbots on the same article happening frequently. At this time there are zero cosbots. -- GreenC 21:51, 8 December 2020 (UTC)

Tasks removed

Hi GreenC. Can you explain why removed without any discussion and real explanation the tasks I submitted for COSDAY? --NicoV ^{(Talk on frwiki)} 18:39, 22 December 2020 (UTC)

These bots are already approved and ran prior to the existence of COSDAY. They are not COSDAY bots. They are not restricted to editing on a 24hr period. COSDAY bots doesn't mean "a bot that makes cosmetic edits", it means "A bot that makes cosmetic edits on a certain day of the month as set forward in the rules at WP:COSDAY". As of this writing, no COSDAY bots have been submitted for approval. -- GreenC 18:44, 22 December 2020 (UTC)

They are in the section Existing tasks, because they are already approved to run but only with other non cosmetic tasks. They are submitted here to be allowed to be run on COSDAY on their own. --NicoV ^{(Talk on frwiki)} 19:16, 22 December 2020 (UTC)

It completely falls in the If you want a COSDAY trial of an existing cosmetic task from an already-approved BRFAs, then list the page below in the Existing tasks subsection.... --NicoV ^{(Talk on frwiki)} 19:24, 22 December 2020 (UTC)

I see sorry I wasn't aware of the already-approved clause. -- GreenC 22:58, 22 December 2020 (UTC)