User talk:ClueBot Commons

This user is NOT a human

This is the combined talk page for ClueBot NG and ClueBot III. These users are automated computer programs and are not humans. Please be aware that bots cannot think like a human and cannot operate outside of their programming. Messages you leave on this talk page will not be answered by a bot – either a bot operator or another human will answer you.

False positives and false negatives

If you believe that ClueBot NG has mistakenly identified a good edit as vandalism, please follow the directions in the warning it gave or click here. Please do not report it on this talk page. It takes less time to report the case to the correct location, and we can handle it more effectively there.
If you believe that ClueBot NG has missed an edit that is vandalism, again do not report it here. ClueBot is unable to catch all vandalism. Just revert the edit and warn the editor.

ClueBot NG Links!

Report False Positives • Review Edits To Improve The ANN  • Frequently Asked Questions

Purpose of this Page

This page is for comments on or questions about the ClueBots.

The current status of ClueBot NG is: Running
The current status of ClueBot III is: Running
Praise should go on the praise page. Barnstars and other awards should go on the awards page.
Use the "new section" button at the top of this page to add a new section. Use the [edit] link above each section to edit that section.
This page is automatically archived by ClueBot III.
The ClueBots' owner or someone else who knows the answer to your question will reply on this page.

Beware! This user's talk page is monitored by talk page watchers. Some of them even talk back.

ClueBot NG Needs You!

Damian has been working hard to bring the Report and Review interfaces back. We are now happy to announce that all is ready!

Report Interface: We need Wikipedians to assist with clearing the backlog of false positive reports

Review Interface: This has a direct effect on the bot and what it knows about vandalism or constructive edits. We need Wikipedians to review edits to effect how the bot should be trained.

Please reply under this message if you want to get involved - Rich^T|C|E-Mail 19:43, 5 August 2025 (UTC)

I'd love to get involved, as a new rollbacker, I wish to help out! Valorrr ^{(lets chat)} 03:07, 6 August 2025 (UTC)

@Valorrr: I've added the relevant rights to you on both interfaces. You'll be able to review edits now - Rich^T|C|E-Mail 11:36, 6 August 2025 (UTC)

@Rich Smith I would be interested in helping out. I just happened to stumble onto this today, as ClueBot kept beating me to several rollbacks I thought I caught instantly! Nubzor [T][C] 20:19, 13 August 2025 (UTC)

@Nubzor: You need to login to either Report or Review first before I can give you the relevant rights, although a new update may have you the rights straight away - Rich^T|C|E-Mail 21:57, 13 August 2025 (UTC)

@Rich Smith Long time no speak! Been busy learning a new job which is why I didn't initially reply. I'm happy to help though. I just logged in and looks like I still have rights from last time 5 albert square (talk) 18:19, 9 October 2025 (UTC)

@5 albert square: Good to hear from you! Perfect, glad to hear it :) - Rich^T|C|E-Mail 11:22, 13 October 2025 (UTC)

@Rich Smith Sorry about that. I didn't even try to log in earlier. I (believe) I was successful at logging into both. I will look into this more tomorrow before I start tinkering. Seems like something more useful/helpful than just patrolling RC. I always chuckle when I am positive a rollback will be successful, only to see it was ClueBot that beat me to it :) Nubzor [T][C] 00:44, 14 August 2025 (UTC)

Quick question for you after taking a look--The Reviewer Interface & instructions provided there make perfect sense. But I'm not 100% certain on the Report Interface. Are we "defer[ing] to reviewer interface" regardless if it's a false positive or not, so that it can then be potentially incorporated into the dataset, either as vandalism or as constructive? Are we just filtering out bugged/invalid entries? Or am I misunderstanding. Nubzor [T][C] 02:18, 14 August 2025 (UTC)

@Nubzor: Essentially yes, if it's an edit that would be helpful for the bot to learn what is/isn't a FP, then defer. If the edit has been suppressed or deleted, then mark it as invalid - Rich^T|C|E-Mail 10:07, 14 August 2025 (UTC)

@Rich Smith Perfect, thank you! Nubzor [T][C] 14:40, 14 August 2025 (UTC)

I'm relatively new to editing but I'd love to help out if I can.

Seantavius (talk) 21:40, 13 August 2025 (UTC)

I see that I've been granted access to the review interface. Thanks

Seantavius (talk) Seantavius (talk) 13:12, 14 August 2025 (UTC)

Sign me up. I'm busy this month but hopefully I'll have more time to dedicate in the coming months —⁠k6ka 🍁 (Talk · Contributions) 19:38, 15 August 2025 (UTC)

@K6ka: You should be able to just sign in, your Wiki rights will give you instant access :) - Rich^T|C|E-Mail 19:43, 15 August 2025 (UTC)

Snazzy snazzy! —⁠k6ka 🍁 (Talk · Contributions) 21:25, 15 August 2025 (UTC)

• @Rich Smith: I can help :-) —usernamekiran (talk) 21:16, 15 August 2025 (UTC)

  @Usernamekiran: Your Wiki rights will automatically give you access upon login - Rich^T|C|E-Mail 11:08, 16 August 2025 (UTC)

  Rich Smith, yes, I logged in, and reviewed some edits successfully. —usernamekiran (talk) 11:38, 16 August 2025 (UTC)

Possible bug: mangling sig embedded in template

Hello. This is a report about a possible bug, in which Cluebot alters a sig embedded in template {{resolved}}, such that it renders improperly on the Archive page, although it looked fine on the Talk page before archiving happened.

At the page Wikipedia:Graphics Lab/Illustration workshop, please note the signature which appears in the comment added in rev. 1316301524 of 17:10, 11 October, 2025 (UTC) in section Special:Permalink/1316301524 § Larger torn table icon; the sig is embedded in the {{resolved}} template, and the wikicode look like this:

Thanks! {{resolved|1=[[User:Mathglot|Mathglot]] ([[User talk:Mathglot|talk]]) 17:10, 11 October 2025 (UTC)}}

and renders like this:

Thanks!

Resolved

 – Mathglot (talk) 17:10, 11 October 2025 (UTC)

So far, so good. After that, there was one subsequent, unrelated edit on the page (rev. 1316434789 of 12:42, 12 October 2025), and then in the next edit, rev. (1316435789 of 12:53, 12 October 2025) Cluebot removed 2 discussions (-4,941 bytes) from Wikipedia:Graphics Lab/Illustration workshop, including the comment with my sig, and added them (+4,946 bytes) to Wikipedia:Graphics Lab/Illustration workshop/Archive/Sep 2025 in rev. 1316435785 of 12:53, 12 October 2025. Note the 5-byte difference: the code line added to the archive appears to have a {{tl}} template introduced to the {{resolved}} template which wasn't there before, thus:

Thanks! {{tl|resolved|1=[[User:Mathglot|Mathglot]] ([[User talk:Mathglot|talk]]) 17:10, 11 October 2025 (UTC)}}

This renders as follows, making the Archive comment sig look garbled, where the prior Talk comment sig was fine:

Thanks! {{[[Template:Mathglot (talk) 17:10, 11 October 2025 (UTC)|Mathglot (talk) 17:10, 11 October 2025 (UTC)]]}}

Is this something that needs to happen for archiving to work properly, and can it be done in a way that does not break the rendered signature post-archiving? Mathglot (talk) 08:54, 13 October 2025 (UTC)

@Mathglot: Huh, that's a bit odd... @DamianZaremba: or @NaomiAmethyst: any ideas what this might be? - Rich^T|C|E-Mail 11:21, 13 October 2025 (UTC)

Housekeeping note: this bug is unresolved, please interpret the 'resolved' badge with checkmark above as part of the bug description, and not of its resolution status. Thanks. Mathglot (talk) 16:08, 22 October 2025 (UTC)

The bot does this explicitly - https://github.com/cluebotng/cluebot3/blob/main/lib/bot.php#L155

The logic has always been there, though I'm not sure what it was intended to achieve.

I can check the logs to see how many times we're applying this (will have to wait a while as I just restarted the bot for different reasons and we don't have good log history are the moment).

I'm pretty hesitant to change the logic as we don't have test coverage and it has quite wide impact with many possible behaviour variants.

- Damian Zaremba ^{(talk • contribs)} 10:44, 29 October 2025 (UTC)

So far I haven't caught any log entries, though I've been working on other things so only checking every now and then. - Damian Zaremba ^{(talk • contribs)} 18:16, 30 October 2025 (UTC)

Thanks all for feedback so far, and pinpointing the trouble spot. Doing it explicitly for some unknown reason is odd; is there anyone still around (or not around—LinkedIn?) we could ask about this? Still monitoring, and hoping this can be made more robust. Mathglot (talk) 18:34, 30 October 2025 (UTC)

User:NaomiAmethyst would be the best bet, though it's been a long time since this was written.

I'll try and have a proper look through the logic and see if we can so something safe, just a little busy at the moment. - Damian Zaremba ^{(talk • contribs)} 22:47, 31 October 2025 (UTC)

No rush, but eyes on it are appreciated. Hopefully something can be done in due course, when the time is right. Thanks, Mathglot (talk) 03:57, 1 November 2025 (UTC)

Feature request: clickable report links in edit summary

The process for reporting a false positive is needlessly cumbersome. A link in the edit summary containing the revert number going to the report page would simplify reporting. With some JavaScript/behind-the-scenes voodo, the user name could be prefilled, too, and the whole process would reduce to click-and-confirm(). Paradoctor (talk) 17:01, 25 October 2025 (UTC)

A link with the revert number prefilled is actually already provided in the level 1 and 2 talk page warnings that the bot provides. That being said, a working link could be added to edit summaries using the "toolforge" interwiki prefix, e.g. .toolforge:cluebotng. I don't know if the interwiki prefix supports adding the revert ID in the link, though; it appears to only work for redirecting to the home page of the tool, so it would only save users a click, rather than prefilling the form. —⁠k6ka 🍁 (Talk · Contributions) 16:35, 29 October 2025 (UTC)

Cluebot NG already parses the query string, adding &user=Paradoctor functionality doesn't seem like a giant leap from there? Paradoctor (talk) 18:44, 29 October 2025 (UTC)

If you are logged in then the user is pre-filled, it's possible that field may become not user-definable, for the same reasons you can no longer anonymously leave comments.

Otherwise feel free to create a pull request at https://github.com/cluebotng/report and it will get reviewed at some point. We should also consider the privacy implications of automatically passing someone's username who may wish to be anonymous.

I only see 2 reports from your username, are you expecting to report a notable amount more? Each report has to be reviewed at least once by an admin, so you could argue that it being slightly cumbersome (requiring some effort from) the reporter is a feature.

- Damian Zaremba ^{(talk • contribs)} 18:15, 30 October 2025 (UTC)

If you are logged in then the user is pre-filled As "Anonymous", not my username. No idea what the rest of that paragraph means.

privacy implications Adding a setting would fix that, no?

are you expecting to report a notable amount more Am I the only one reporting? I don't know what a "notable" amount is, but yes, you can expect more reports from me, and presumably from others.

Each report has to be reviewed at least once by an admin That's the job. Are you implying reporting is not a good thing? More to the point, reporting should lead to less false positives, and therefore to less work for both admins and editors. Right? Paradoctor (talk) 19:16, 30 October 2025 (UTC)

> As "Anonymous", not my username.

I don't see your user account in report which would imply you haven't logged in, the code is quite simple for setting the current username, so I would be surprised if it's broken.

> Adding a setting would fix that, no?

Where would the setting exist?

What would the default be?

How would it be communicated to users?

> Am I the only one reporting?

If you are only reporting once or twice a small amount of overhead isn't a large issue, if you are reporting hundreds of times then there are likely other questions in play.

Bear in mind you're talking about things that are used by many many people, but actively maintained by 1-2. Every single feature and variant increases maintainer overhead and in general we're trying to remove rather than add to them.

> That's the job

Feel free to volunteer for it.

> Are you implying reporting is not a good thing?

I'm implying that forcing someone to follow a well defined process has side effects relating to reports that are pure spam/abuse.

In general this is not an uncommon practice within Wikipedia/Wikipedia.

> More to the point, reporting should lead to less false positives, and therefore to less work for both admins and editors.

That is the theoretical goal, though given we are about 14 years since the last data set was trained, those 2 things are very far apart.

As I said before, feel free to contribute a PR if you think it's a valuable thing. Otherwise don't expect random things to always be actioned, ideas are good but also easy, without general consensus or wider reasonings it's unlikely to be actioned by a maintainer. - Damian Zaremba ^{(talk • contribs)} 22:16, 31 October 2025 (UTC)

A barnstar for you!

	The Anti-Vandalism Barnstar
	Best antivandal bot ever! Thanks Bryce M (talk) 11:49, 26 October 2025 (UTC)

I don't understand

ClueBot has undone some of my edits on redirects to [[List of Star Wars Legends characters]] and calls them vandalism. Could someone explain me why ?

Thank you Je peux écrire maintenant ? (talk) 09:46, 29 October 2025 (UTC)

Je peux écrire maintenant ?, what you experienced is called a "false positive". ClueBot NG is an anti-vandalism bot, meaning that it automatically tries to determine which edits are vandalism and which aren't, and then reverts edits that it thinks are vandalism. Sometimes, the bot makes a mistake and reverts an edit that isn't vandalism (think of how legitimate emails sometimes end up in your spam or junk mail folder). If that happens, you can simply undo ClueBot NG's edit, and then report the false positive. —⁠k6ka 🍁 (Talk · Contributions) 16:30, 29 October 2025 (UTC)

Okay, thanks. Je peux écrire maintenant ? (talk) 20:15, 29 October 2025 (UTC)

Request for archive of my user talk page in Sindhi Wikipedia

Hello, I would like to request you to archive my user talk page on Sindhi Wikipedia. I'm looking forward to hearing from you soon. Warm regards. JogiAsad (talk) 21:26, 31 October 2025 (UTC)

Hello,

Have you read https://en.wikipedia.org/wiki/User:ClueBot_III/Documentation?

There are numerous options for configuring the bot, so generally it is up to the user to configure (otherwise we're just guessing what your requirements are).

If you can expand on what you would like to achieve (retention, structure etc) then maybe someone can propose a template block to use. - Damian Zaremba ^{(talk • contribs)} 22:52, 31 October 2025 (UTC)

False positives quite high?

Does anyone else feel Cluebot is reverting too many legitimate edits? The bot feels like its doing more harm than good at this point.

From my watchlist we have [1] reverting a spelling fix and reverting a perfectly good-faith addition [2], during this same watch period it only reverted two vandals. I do report the false positives but it is just getting tiresome to spend more time reporting false positives than the time saved by the bot reverting obvious vandalism. Traumnovelle (talk) 21:31, 4 November 2025 (UTC)

Might be a bug with the new temporary accounts that got rolled out? Definitely something to look into. — Naomi Amethyst 22:53, 4 November 2025 (UTC)

I noticed this before temporary accounts were a thing. I'm unsure if I'm just simply 'unlucky' or if there is a boarder pattern of an issue with the artificial intelligence here. Traumnovelle (talk) 01:35, 5 November 2025 (UTC)

Any chance you can say form when you noticed it?

The amount of reverts has increased in the last couple of months [3].

There where 2 notable changes made to the bot in august, 1 to re-try the replica credentials multiple times (should have increased number of processed edits) and 1 to use REDIS rather than a file for storing recently reverted titles (affects re-reverts, previously where lost on every restart).

There was a bug related to (Huggle) whitelisted users for a few weeks, which was fixed a couple of weeks ago.

Additionally there was some changes regarding logging, container resources, PHP runtime, monitoring [4] etc.

Part of those where to make the bot skip less edits, which could also result in more false positives (but at the same ratio as before).

- Damian Zaremba ^{(talk • contribs)} 15:32, 5 November 2025 (UTC)

@DamianZaremba I think this is closely tied to the deployment of temporary accounts - yesterday it looks like ClueBot did ~2x the number of reverts it usually does: https://quarry.wmcloud.org/query/98804 Sam Walton (talk) 15:54, 5 November 2025 (UTC)

Thanks, that's useful info.

I pulled one of the edits above into review [5], core is scoring that under the threshold, but it was over at the time it was reverted... which to be fair could be a review bug, but could also be something funky in the xml.

Looking at some recent anon (temp account) reverts by eye a few do stand out as suspicious. Having a dig through the logs we have at the moment.

- Damian Zaremba ^{(talk • contribs)} - Damian Zaremba ^{(talk • contribs)} 16:41, 5 November 2025 (UTC)

Had a quick look at what's being scored and everything looks reasonable.

The main thing that comes to mind, is now every anon user has a user_edit_count & user_distinct_pages of 0 rather than from what was previously made on the IP (and a registration time of something that is not the time of the edit).

Those do affect the score generated by core, which would be a potentially large change for users behind large ISPs (lot's of usage of CGNAT addresses).

Given a vandal can now just drop their cookies to get a new temp account, that might change how the bot needs to handle things more generally.

Need to find a clean set of edits that are clearly false positives to test some theories I think. - Damian Zaremba ^{(talk • contribs)} 17:17, 5 November 2025 (UTC)

Hm I don't think I can give a precise date but maybe since around this past month? So perhaps temporary accounts are causing an issue too but I noticed this before they were rolled out. Traumnovelle (talk) 18:51, 5 November 2025 (UTC)

After digging into 1320422031 from above, I've pushed a fix for the 'recent window', which affects the page recent edit/revert counts (has been there a good while).

With the previous behaviour core scores at 0.904905, now it scores at 0.7882, which drops it below the revert threshold, which would have skipped the 2 example changes above.

The fix for (Huggle) whitelisting is also now deployed.

I don't see a wider issue at the moment, but will have a look again tomorrow.

- Damian Zaremba ^{(talk • contribs)} 01:59, 6 November 2025 (UTC)

Thank you. I'll keep reporting false positives through the panel for now but hopefully this helps to reduce most of them. Traumnovelle (talk) 02:14, 6 November 2025 (UTC)

Looks like that had some impact, we're at about 400 edits in 14 hours, vs about 700 on the 4th (there was an outage overnight on the 5th, so graphs are funky). Hour by hour based on the replica's also show roughly half (lies, damn lies and statistics etc).

This was a very long standing bug, so I suspect temp accounts dropping the "editor's" previous changes combined with inflated "recent" page edits affected the score noticeable.

Will have a look in a couple of days, but for now I think things are roughly back on track.

- Damian Zaremba ^{(talk • contribs)} 14:41, 6 November 2025 (UTC)

We're back to around 300 edits on Main for the last few days, so unless anyone screams, I'll class this as resolved.

- Damian Zaremba ^{(talk • contribs)} 19:03, 9 November 2025 (UTC)

I haven't noticed any issues as of late myself. Traumnovelle (talk) 19:17, 9 November 2025 (UTC)

Thanks for confirming.

I've roughly added affected edits (if a 're-score' is below the revert threshold) to the review interface and created any reports that have not already been made.

That's added quite a bit to the backlog, but should make sure this "bad data" doesn't come back to bite us in the future.

They are all attached to a new edit group [6] so we have some reference outside of general reports.

- Damian Zaremba ^{(talk • contribs)} 20:35, 9 November 2025 (UTC)

• This bot is not ready for prime time. ~2025-32857-32 (talk) 15:27, 11 November 2025 (UTC)

  Feel free to file an issue [7] describing how you would like to improve it.

  - Damian Zaremba ^{(talk • contribs)} 18:49, 11 November 2025 (UTC)

Here

Starlet147 has given you microchips! Microchips promote WikiLove (📖💞) and hopefully this one has made your day more efficient. It is the food best preferred by bots. 🤖 Spread the WikiLove by giving someone else microchips, whether it be someone you have had robot wars with in the past or a good friend.

Spread the goodness of microchips by adding {{subst:Microchips for you}} to someone's talk page with a friendly message!

Starry~~(Starlet147) 22:41, 4 November 2025 (UTC)

Delicious Motor Oil for you

BumblB645 has given you motor oil! Motor oil promotes WikiLove (📖💞) and hopefully this one has made your day more efficient. It is the drink best preferred by bots. 🤖 Spread the WikiLove by giving someone else motor oil, whether it be someone you have had robot wars with in the past or a good friend.

Spread the goodness of motor oil by adding {{subst:Motor oil for you}} to someone's talk page with a friendly message!

BumblB645 (talk) 23:36, 6 November 2025 (UTC)