"Already has a bot flag(Y/N): Y (But it wont use it, all of its functions should be in watch-lists)"
I hope this is some horrible typo, watch-lists have nothing to do with bot flags...
Also, per the Dryrun, this and this would of been reverted. I know it's great to detect a large amount of vandalism, but being too sensitive and having false positives is not an acceptable side affect. This is present in ClueBot and other antivandalism bots. FinalRapture - †☪ 21:18, 8 June 2010 (UTC)
Both those false positives have been fixed, even before I read you comments. Finding errors like these were the whole purpose of the dry run.
The source is not yet available because right now it looks like a 5 year old wrote it, Im cleaning it up now, and will publish it after the full trial.
It won't edit with a bot flag because its edits should not appear in the recent changes feed. Sorry, it was early when I filled this all out. Tim1357talk 23:47, 8 June 2010 (UTC)
What are the problems you have with running this under the same bot account? Its easier for me to do, and I dont think it makes it any harder for others to use, given the propper documentation and logging. Tim1357talk 23:47, 8 June 2010 (UTC)
"It won't edit with a bot flag because its edits should not appear in the recent changes feed.". What the hell? FinalRapture - †☪ 02:07, 9 June 2010 (UTC)
┌────────────────────────────────────────────────────────────────────────────────────────────────────┘If I can quote WP:BOTPOL:
Edits by such accounts are hidden by default within recent changes.
Sorry if I was unclear. Tim1357talk 01:10, 10 June 2010 (UTC)
The comment about the watchlist isn't as ridiculous as FinalRapture makes it, since there is a user preference that enables you to hide bot edits from your watchlist. Then again, you can't run one task with a bot flag and another without one on the same account. Ucucha 16:54, 11 June 2010 (UTC)
My bad, thanks for teaching me something new. Ucucha 06:15, 13 June 2010 (UTC)
Compliance with 1RR
I'v put in some thought on how to make this bot not revert the same edit over and over again. The process I've come up with is this:
When the bot reverts an edit, it makes a hash (to save memory) of the username, the page, and the rendered diff. This hash is then stored, along with a time stamp.
If within 24 hours, another edit has an identical hash (meaning it is the same user, making the same edit on the same page) the bot will make the revert_score threshold lower. Instead of needing a score of -4 to be reverted, it would need a score of -10 (or something like that) . This means extremely blatant vandalism will be reverted again and again. However, this feature can be turned off
IMO AVBots should be 1RR. It's better then picking arbitrary numbers as a cutoff. QTC 01:39, 13 June 2010 (UTC)
So even though an edit has a score of -1,000, the bot should not re-revert? Or are you saying the bot should not revert the same user in the same day, even if it is a different page/edit? Tim1357talk 01:45, 13 June 2010 (UTC)
After a brief discussion on IRC, I have compromised so that the hash includes the user and the page title, so that the bot will never revert two edits the same page by the same user within 24 hours of each other. Tim1357talk 02:31, 13 June 2010 (UTC)
Yes, but the payoff is that other users can help build the regex base. Additionally, I will request that the page be fully protected, so that only admins may edit it. I believe that admins will be smart enough to not modify a regex if they do not know how. I might be overestimating them though. Tim1357talk 21:14, 13 June 2010 (UTC)
Yes, I have been considering making the page a redirect to a .css page of mine. In fact, that's just what I'll do. Tim1357talk 21:33, 13 June 2010 (UTC)
Addition: "if the user is not in Huggle's whitelist", does that mean that the bot can revert some autoconfirmed users? Sole Soul (talk) 19:31, 13 June 2010 (UTC)
Are there auto confirmed users that aren't in the white-list? Tim1357talk 21:14, 13 June 2010 (UTC)
I don't know, I don't use Huggle :) Sole Soul (talk) 21:26, 13 June 2010 (UTC)
Apparently not, "Huggle whitelists users with edit counts above 500". Autoconfirmed users should not be reverted by a bot. Note: 3 of the 6 false positive edits reported were made by autoconfirmed users. Sole Soul (talk) 22:30, 13 June 2010 (UTC)
If it's alright with you, I'd like to stick with the Huggle white list. I spent some time looking, and it appears there is nowhere that the API will let me download a list of autoconfirmed users. In fact, the database does not even have a place where it saves a list of these users. Hopefully the bot is coded well enough so that this will not effect performance. Tim1357talk 03:32, 14 June 2010 (UTC)
Of course if there is a technical limitation, you have no choice, but I wonder how Clubot and AVBOT handled the situation, I'm not sure. Sole Soul (talk) 07:51, 14 June 2010 (UTC)
┌────────────────────────────────────────────────────────────────────────────────────────────────────┘Let me look at AVBOT's code. Tim1357talk 13:56, 14 June 2010 (UTC)
ClueBot doesn't revert users with > 50 edits, or IPs with > 500 edits. (X! · talk) · @219 · 04:15, 15 June 2010 (UTC)
Do you know if it loads that entire list all at once, or it checks for each edit? Tim1357talk 04:54, 15 June 2010 (UTC)
Each edit. (X! · talk) · @248 · 04:56, 15 June 2010 (UTC)
I think all IP edits should be checked, as school IPs can accumulate large edit numbers. Sole Soul (talk) 11:06, 15 June 2010 (UTC)
I agree. The bot will ignore edits by non-IP users that have more than 50 edits. The bot still uses Huggle's whitelist, because that list is helpful to strip bots, admins, and other experienced editors before it downloads the edit. It checks the user's edit count after it downloads and evaluates the edit.Tim1357talk 16:17, 15 June 2010 (UTC)
Oh yeah, I finally got a github, and I've been updating the source here. Tim1357talk 16:17, 15 June 2010 (UTC)
Trial or approval at this point? I could go for a trial. MBisanztalk 18:50, 26 June 2010 (UTC)
┌────────────────────────────────────────────────────────────────────────────────────────────────────┘Trial plz. Tim1357talk 00:34, 27 June 2010 (UTC)
Approved for trial (50 edits). MBisanztalk 04:36, 27 June 2010 (UTC)
Trial complete.. 50 Reversions. There were a few false positives, and all were related to one regex. That particular regex has sine been removed. There were a few other bugs, but all were easy to fix. Tim1357talk 02:06, 28 June 2010 (UTC)
Approved for trial (20 edits). As you requested. (X! · talk) · @133 · 02:11, 28 June 2010 (UTC)
Done, along with a bit extra monitored sessions. No errors were encountered. I suggest either approval or another long trial period (a week or so). Tim1357talk 17:07, 4 August 2010 (UTC)