Wikipedia:Bots/Requests for approval/SoxBot III 3
- The following discussion is an archived debate. Please do not modify it. Subsequent comments should be made in a new section. The result of the discussion was Approved.
Contents
Operator: Xclamation point
Automatic or Manually Assisted: Automatic
Programming Language(s): PHP
Function Summary: Anti testing bot
Edit period(s) (e.g. Continuous, daily, one time run): Continuous
Already has a bot flag (Y/N): Y, old tasks inactive
Function Details: This bot is based off of ClueBot, in the fact that it has heuristics and has a score function to rate tests. This bot is designed to combat edits which add stuff like "<math>Insert formula here</math>== Headline text =='''Bold text'''''''Italic text''''". It sits in the RC feed, and if it detects an edit by an IP with less than 250 edits, or a user with less than 50 edits, it checks the diff. It checks the parts added and removed against a score list, and a whole list of common testing techniques are scored depending on how much of a false positive it could be. If it falls below a certain range, the bot reverts the edit.
Discussion
editAn IP with less than 250 edits? That's probably 90% of anons. Probably most of the ones with more are shared IPs. If the threshold is that high, it would probably be more efficient to just check every IP edit. Also, will it be limited to checking mainspace pages? Mr.Z-man 06:59, 28 December 2008 (UTC)[reply]
- I could make it all IPs, I guess. And yes, it is limted to Mainspace. Xclamation point 07:00, 28 December 2008 (UTC)[reply]
What will it do if someone makes an otherwise legitimate edit where they accidentally include a string such as '''Bold text'''? rspεεr (talk) 09:54, 28 December 2008 (UTC)[reply]
- In that case, they can make the edit again, and the bot won't revert it. Xclamation point 18:52, 28 December 2008 (UTC)[reply]
- In addition, it will not revert if the edit adds more than 1000 characters, unless the score is unusually high. Xclamation point 06:54, 30 December 2008 (UTC)[reply]
- Sure. It sounds like the score function deals with the cases I was worried about. rspεεr (talk) 08:13, 30 December 2008 (UTC)[reply]
- In addition, it will not revert if the edit adds more than 1000 characters, unless the score is unusually high. Xclamation point 06:54, 30 December 2008 (UTC)[reply]
I could have sworn there was a bot already approved for this just recently... --MZMcBride (talk) 09:55, 28 December 2008 (UTC)[reply]
- Ameliorationbot? Yes, but Ameliorate has just retired. Xclamation point 18:52, 28 December 2008 (UTC)[reply]
Seems to be making good reversions. But as of 1500GMT today (Jan 2) it seems to have stopped issuing warnings to users. Martin 22:05, 2 January 2009 (UTC)[reply]
Short dry run
editFrom a short dry run, here's what the bot would have reverted:
- Would revert revision 260557092 on page Religion in Iraq: Registers a score of -24.
- Would revert revision 260618150 on page Mille-feuille: Registers a score of -44.
- Would revert revision 260619225 on page National Institute of Technology, Rourkela: Registers a score of -49.
- Would revert revision 260633058 on page Lal Jose: Registers a score of -20.
- Would revert revision 260633850 on page Idaho Vandals: Registers a score of -55.
- Would revert revision 260636449 on page Machangulo: Registers a score of -20.
- Would revert revision 260638078 on page Microsoft Word: Registers a score of -101.
- Would revert revision 260643445 on page Marion Barber III: Registers a score of -24.
- Would revert revision 260675316 on page Tryon Palace: Registers a score of -24.
- Would revert revision 260716745 on page London Dreams: Registers a score of -20.
- Would revert revision 260735549 on page HOMO/LUMO: Registers a score of -16.
- Would revert revision 260770751 on page Melting: Registers a score of -29.
- Would revert revision 260821527 on page Rye, Victoria: Registers a score of -3.
- Would revert revision 260822353 on page Eleca: Registers a score of -5.
- Would revert revision 260822483 on page Eleca: Registers a score of -5.
- Would revert revision 260822592 on page Dear Mr. Henshaw: Registers a score of -33.
- Would revert revision 260822638 on page Craig Zielinski: Registers a score of -5.
- Would revert revision 260825705 on page Jerrod Johnson: Registers a score of -5.
- Would revert revision 260826847 on page Voyages of Christopher Columbus: Registers a score of -18.
- Would revert revision 260827148 on page H. H. Holmes: Registers a score of -6.
- Would revert revision 260828637 on page KGB: Registers a score of -5.
Xclamation point 07:49, 29 December 2008 (UTC)[reply]
- Update: The results above are from an overnight dry run. Xclamation point 15:31, 29 December 2008 (UTC)[reply]
I have a few questions. First, at least one of the previous bots that did this was Wikipedia:Bots/Requests for approval/TestEditBot (which has a better name at least...). There seem to be a lot missing safeguards here (or at least they're not described).
- Does it give user warnings to users making test edits, and if so, which templates will it use?
- It will give warnings
, but I have not written the page which will warn as of yet.see below. Xclamation point 20:05, 29 December 2008 (UTC)[reply]
- It will give warnings
- If you're using ClueBot's classes and ClueBot is (as far as I'm aware) still running, why wouldn't this functionality be built into that bot?
- Because it has a whole different mindset on what to do with a test edit than ClueBot does for vandalism. ClueBot is more of "angry reversions, report it to AIV, and everywhere else", while this bot is more of a "revert and warn nicely" approach (That's my view on it, anyway). Xclamation point 20:05, 29 December 2008 (UTC)[reply]
- How does it check the number of edits for a given IP?
- It doesn't anymore, per Z-Man's comment above. Before, it used the API. Xclamation point 20:05, 29 December 2008 (UTC)[reply]
- Further, is the number of edits for an IP relevant to whether test code was added, and if so, how?
- See above question Xclamation point 20:05, 29 December 2008 (UTC)[reply]
- You vaguely mention that this bot won't revert the same person twice on the same day. Is this source code publicly available and has it been checked by others?
The source is not publicly available, as of yet. Right now, the code is pretty ugly looking. I will work on making it cleaner code, and then release it publicly, and have it reviewed.Xclamation point 20:05, 29 December 2008 (UTC)[reply]- I have released the code at User:SoxBot III/Source, and
it is currently beinghas been reviewed by Chris G. Xclamation point 03:25, 30 December 2008 (UTC)[reply]
- I have released the code at User:SoxBot III/Source, and
- Going beyond that, how much of the code for this bot was written by you (Soxred)? As far as I'm aware, Cobi is pretty inactive, so if something breaks, will you be able to fix it? Or was this code simply re-used (not that there's anything wrong with that) from previous bots? And if it is re-used code from previous test edit detection bots, why has this project been abandoned previously? (This also sort of ties in to why ClueBot isn't doing this itself, though I have my suspicions.)
- Maybe 60% was written by me. The other 40% is the IRC bot part (for getting the live feed, and also for getting a machine readable diff. I understand the concern about Cobi being inactive, but if it breaks, it most likely will be easy to fix by myself. It is also not using any code from previous test detection bots. Xclamation point 20:05, 29 December 2008 (UTC)[reply]
Thanks! --MZMcBride (talk) 19:27, 29 December 2008 (UTC)[reply]
This looks pretty good. I take it this is just for the test edit type things and not the other types of vandalism scoring that cluebot does (various words etc)? I'm not much for reading code, but the tests I saw were of the '/\'\'\'Bold text\'\'\'/' type. Either way this is great since a fair number of these slip through and the more a bot can get the better. Another thought I had, what's the chance of checking if a user is doing this repeatedly and reporting them? Anybody doing the same thing after being warned for example would clearly be an issue. - Taxman Talk 19:18, 1 January 2009 (UTC)[reply]
- I suppose I could make it report to AIV if they've been warned 4 times. Would that work, Taxman? Xclamation point 19:41, 1 January 2009 (UTC)[reply]
- For IP's I think that would be about right, though for registered users I don't see why more than two or three is needed. If the bot really warned in error because someone added the mistakes unintentionally, a report shouldn't be a problem, because it can be explained on their talkpage or whatever and they wouldn't get blocked. Whoever reviews the report will look into it. But by two or three false warnings someone really needs to be reminded to use preview or show changes. :) two or three real warnings they need a block. But for IP's who knows who is behind it so unless the warnings are several or over a short period, no report is likely needed, justifying the higher number. - Taxman Talk 20:18, 1 January 2009 (UTC)[reply]
Warning templates
edit== December 2008 ==
Welcome, and thank you for experimenting with Wikipedia. Your test on the page [[:{{{1}}}]] worked, and it has been automatically reverted. Please take a look at the welcome page to learn more about contributing to this encyclopedia. If you would like to experiment further, please use the sandbox. If you believe there has been a mistake and would like to report a false positive, please report it here, but be sure to mention this number: {{{2}}}. Thank you. SoxBot III (talk | owner) 06:52, 30 December 2008 (UTC)[reply]
Please stop making test edits to Wikipedia, as you did to [[:{{{1}}}]]. It is considered vandalism, which, under Wikipedia policy, can lead to blocking of editing privileges. If you would like to experiment again, please use the sandbox. If you believe there has been a mistake and would like to report a false positive, please report it here, but be sure to mention this number: {{{2}}}. Thank you. SoxBot III (talk | owner) 06:52, 30 December 2008 (UTC)[reply]
This is the last warning you will receive for your disruptive edits.
The next time you disrupt Wikipedia, as you did to [[:{{{1}}}]], you will be blocked from editing. If you believe there has been a mistake and would like to report a false positive, please report it here, but be sure to mention this number: {{{2}}}. Thank you. SoxBot III (talk | owner) 06:52, 30 December 2008 (UTC)[reply]
- Approved for trial (1 week). Please provide a link to the relevant contributions and/or diffs when the trial is complete. --Chris 04:22, 31 December 2008 (UTC)[reply]
- Trial started - After a full day of trying to get it work, with server hopping and the toolserver being down, I FINALLY got it to work. It is now reverting and warning users. Xclamation point 04:34, 1 January 2009 (UTC)[reply]
- {{BotTrialComplete}} - I have also changed it to simply undo the edit, rather than rollback. I think that will cause less false positives. A report of false positives can be found here. Xclamation point 21:02, 6 January 2009 (UTC)[reply]
- Trial started - After a full day of trying to get it work, with server hopping and the toolserver being down, I FINALLY got it to work. It is now reverting and warning users. Xclamation point 04:34, 1 January 2009 (UTC)[reply]
Additional discussion
editI noticed that the false positives were caused by a user entering legitimate text and then something like ''Italic text''. Of course vandals also often enter such strings in their crap edits. Based on your response to the one FP, I'm guessing the bot will no longer make any changes if a certain threshold of "non-test" text is also entered. Perhaps, in these cases the bot could just remove the "test" portion (probably without warning), but only when no text was deleted in the process. But perhaps not... the downside would be potentially vandalious edits being only partially reverted. Just something to consider. --ThaddeusB (talk) 23:06, 6 January 2009 (UTC)[reply]
- I have lowered the point value for those strings. Xclamation point 04:19, 10 January 2009 (UTC)[reply]
Extended trial
editApproved for trial (14 days). Please provide a link to the relevant contributions and/or diffs when the trial is complete. BJTalk 02:17, 15 January 2009 (UTC)[reply]
- Trial complete. During the trial, I got 2 or 3 complaints about false positives. Those are likely not to happen again, as I have raised the threshold to eliminate those specific false positives. Xclamation point 00:30, 29 January 2009 (UTC)[reply]
- Approved.Everything looks good, I see no reason not to approve. Richard0612 21:22, 30 January 2009 (UTC)[reply]
- The above discussion is preserved as an archive of the debate. Please do not modify it. Subsequent comments should be made in a new section.