Wikipedia

Wikipedia:Bots/Requests for approval/DefconBot

< Wikipedia:Bots | Requests for approval
This is an old revision of this page, as edited by A930913 (talk | contribs) at 12:13, 16 May 2014 (→DefconBot: +reply). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

DefconBot

Operator: A930913 (talk · contribs · SUL · edit count · logs · page moves · block log · rights log · ANI search)

Time filed: 15:29, Wednesday May 14, 2014 (UTC)

Automatic, Supervised, or Manual: Automatic

Programming language(s): Edits performed by pywikibot.

Source code available: On the labs. (/data/project/defconbot)

Function overview: VoxelBot is dead. Long live DefconBot!

Links to relevant discussions (where appropriate):

Edit period(s): On the half and hour. Calculated chance of defcon change = ~30% giving an effective rate of once per hour and a half.

Estimated number of pages affected: 1

Exclusion compliant (Yes/No): No

Already has a bot flag (Yes/No): No

Function details: A new mathematical approach to automatically updating the defcon. There is already the precedent set by VoxelBot for updating the defcon automatically, so the discussion is mainly about the new metric.

ClueBotNG not only reverts vandalism, but also scores every edit with its ANN. I have recorded over the period of a month the sum of these scores every minute. [1] I have dubbed this metric the ClueBot Vandal Score (CVS). The CVS, at any one time, gives the approximate quantity of vandalism currently being added to the wiki. One can take an average (or sum, for it matters not which) of these scores over a period of time, for instance half an hour, to work out a score for said period of time.

Using the aforementioned history collected, we can distribute a percentage of the time we want a particular defcon to be spent on and calculate the thresholds required to achieve those percentages. I have done some graphing and these percentages look good.

Defcon Level Percentage of time Triggered
1 1% Monday & Tuesday peak
2 10% Weekday peak
3 20% Weekend peak
4 60%
5 9% Daily minimum

The bot is currently updating the template User:DefconBot/defcon
The current values are:

  • level = 0
  • info = Cannot determine vandalism level according to DefconBot
  • sign = 05:31, 2 September 2016 (UTC)

A live visualisation of the vandalism stats is at //tools.wmflabs.org/cluestuff/stats.html

I have spent a long time trying to make a mathematical formula to include the complexities of vandals vs counters, but I have come to the realisation that current metric proposed, however simple, is the best one. Other metrics can be updated or added with further requests. 930913(Congratulate) 15:29, 14 May 2014 (UTC)[reply]

Discussion

  •   Note: This bot appears to have edited since this BRFA was filed. Bots may not edit outside their own or their operator's userspace unless approved or approved for trial. AnomieBOT⚡ 16:02, 14 May 2014 (UTC)[reply]
Derp.   Fixed 930913(Congratulate) 20:36, 14 May 2014 (UTC)[reply]

Pinging people who have recently updated the template manually: K6ka, Novusuna, Josve05a, TheMesquito, BethNaught, Eyesnore, Slazenger & Jeffrd10. 930913 {{ping}} 21:43, 14 May 2014 (UTC)[reply]

...so, does this mean I lose my job? :( --k6ka (talk | contribs) 22:16, 14 May 2014 (UTC)[reply]
It seems to me that measuring the rate of vandalism is better than the rate of reversion. I'd be happy to see this have a test run - if the techy people who know what they're doing think it's ok. BethNaught (talk) 06:41, 15 May 2014 (UTC)[reply]
So would I. I'm in support of this new bot, as it'll hopefully return more accurate results than just how much vandalism is being reverted (the more vandalism reverted, the better. That doesn't really help much as it pretty much proves that the anti-vandalism team on duty is able to handle things or whatnot). I do have concerns for this template when ClueBot NG goes down, as it does often thanks to WMF Labs... what will this template display if CBNG data is not available? --k6ka (talk | contribs) 10:55, 15 May 2014 (UTC)[reply]
I'd also like to see this get a trial run; this seems to be a better method of measuring vandalism than reverts per minute. I share K6ka's concerns about what happens when ClueBot goes offline. Some kind of fallback is needed, maybe even something as simple as setting the defcon to 1 with a note that CBNG is down. Novusuna talk 17:21, 15 May 2014 (UTC)[reply]

@K6ka: It means you've been promoted  
So I've made a separate script that will check if the CBNG feed is working and if not, fail into a level 0 update. (Yes, level 0 exists.) So, failure scenarios:

  1. CBNG goes down: No edit is made on the half hour, though one minute later, the failedit is made. When CBNG comes back up, the first edit may be based on old data.
  2. CBNG is up, but the feed is down: As #1
  3. CBNG or the feed messes up and slowly trickles partial information through: May give false (lesser) levels, but if less than 1/min, the failedit should fire.
  4. Main DefconBot script fails: No edits.
  5. Labs fails: No edits.

930913 {{ping}} 23:42, 15 May 2014 (UTC)[reply]

@A930913: Well, seeing that WMF Labs causes a lot of problems (such as disabling ClueBot NG and making bots edit whilst logged out, tripping edit filters from head to toe while recent changes patrollers watch bewildered, scratching their heads). While I know that a level 0 existed on the template, I'm not actually sure how to get it to show. It would be useful for me (and anyone reading this) to know so in case WMF does screw up, I can edit the template with level 0 manually. --k6ka (talk | contribs) 03:08, 16 May 2014 (UTC)[reply]
@K6ka: Like the good old days. Though a human is usually capable of working out a non zero level. (Even if it is a lv1 "OMG ClueBot is down".) 930913 {{ping}} 12:13, 16 May 2014 (UTC)[reply]
Retrieved from "https://en.wikipedia.org/w/index.php?title=Wikipedia:Bots/Requests_for_approval/DefconBot&oldid=608824704"