Wikipedia talk:STiki/Archive 4



Minor flag still on for good faith edits?
I notice the minor flag is still on for reversions of "good faith edits". Is this intentional? If so, could an option be put in to turn this flag off? Thanks! Allens (talk &#124; contribs) 19:09, 11 April 2012 (UTC)


 * Acknowledged. I was just passing the "good faith reverts" through the "vandalism revert" system with instructions not to warn. Obviously somewhere in there the "minor edit" flag has become a hard-coded default. I will fix the issue and file as T#006. Thanks, West.andrew.g (talk) 19:15, 11 April 2012 (UTC)


 * Is that flag automatic for use of the native "rollback" among those with that permission? Allens (talk &#124; contribs) 19:42, 11 April 2012 (UTC)


 * Whether its automatic or not, I am unsure. However, I do know that via the API that "minor" flag of rollback can be set, so this is not an issue. Thanks, West.andrew.g (talk) 19:54, 11 April 2012 (UTC)


 * I lied. Per the API all rollback edits will be labeled as "minor" with no option to override. How do we handle this? Simulating rollback with standard-edit functionality is possible; but not good in terms of bandwidth and system resources. Reading WP:Minor, our AGF cases are likely to be "near-vandalism" or "borderline-vandalism" which makes them close to qualifying as minor? I am pulling the feature request until we flesh this out better. West.andrew.g (talk) 20:10, 11 April 2012 (UTC)


 * I am concerned by: "Reverting a page is not likely to be considered minor under most circumstances. When the status of a page is disputed, and particularly if an edit war is brewing, then it is better not to mark any edit as minor. Reverting blatant vandalism is an exception to this rule." I note the word "blatant". Perhaps simulating rollback (done by such tools as Twinkle anyway...) and asking the MediaWiki developers for an API modification is indicated? Allens (talk &#124; contribs) 20:22, 11 April 2012 (UTC)


 * Hmmmm... Still a borderline case I believe. When we are talking about AGF reverts there certainly isn't an implied "dispute" or "edit war" going on. We have a "vandalism" button for "blatant vandalism", and I would say that while AGF reverts aren't necessarily vandalism they are "blatantly unconstructive." The point of minor edits is for watchlist monitoring, so watchlisters don't waste their time reviewing "blatant" actions. In this vain, I would say we are in the spirit of the rule by marking AGF reverts as minor. After all, STiki users aren't topic experts, they are just classifying whatever the queue pops them -- thus if they can see something wrong it must be pretty obvious in nature. Regardless, I've posted over at the Technical Village Pump for advice. Thanks, West.andrew.g (talk) 22:22, 11 April 2012 (UTC)


 * As promised, posted to the Village pump (technical) for discussion. Thanks, West.andrew.g (talk) 22:35, 11 April 2012 (UTC)


 * It was easier to hack out than I had imagined. Per the just released 2012-04-12 version, all good-faith reverts will NOT be marked as "minor". Thanks, West.andrew.g (talk) 03:13, 12 April 2012 (UTC)


 * I am delighted by the introduction of the AGF option. Sometimes I felt a little harsh reverting certain untidy / non-constructive edits under a pure vandalism banner, so this is significant boost IMO. Orphan Wiki (talk) 10:41, 12 April 2012 (UTC)

Blank diff?
STiki is showing a blank page for the Phineas and Ferb most recent edit + 1 back (as rollback). I think this may be the equivalent of it earlier showing no visible differences; I had thought it was due to spacing differences, but now I have to wonder. Allens (talk &#124; contribs) 19:41, 11 April 2012 (UTC)


 * This is not an error. Blank diffs will be shown if a user self-reverts their own contributions (as is the case here) -- this is caused by the diff browser now displaying rollback derived differences. I would say these should always be classified "innocent", and in the future I may write some logic to toss these out automatically. Until then, though, all is fine. Thanks, West.andrew.g (talk) 19:58, 11 April 2012 (UTC)

Low-importance bug: Save of "External Links" setting doesn't work
It saves being checked off, but not actually displaying the links. Allens (talk &#124; contribs) 21:54, 11 April 2012 (UTC)


 * I found this error to be reproducible, filed as T#007. Thanks, West.andrew.g (talk) 22:09, 11 April 2012 (UTC)


 * Error fixed per the release of 2012-04-12. Ticket T#007 wiped from active table. Thanks, West.andrew.g (talk) 03:14, 12 April 2012 (UTC)

Bug fixes by meiskam
Andrew, Were all the bug fixes by meiskam, which he mentioned in this post included in the latest version of STiki? Yaris678 (talk) 12:58, 12 April 2012 (UTC)
 * Not all of my commits were included. Sounds like we're going to wait for the translation of the gui until code has been added so that the tool works on other language versions of Wikipedia as well. –meiskam (talk•contrib) 14:04, 12 April 2012 (UTC)
 * Does that mean all of the straight bug fixes were included but the language-related aspects were not? That would make sense... but I am just checking.
 * BTW. I think having STiki in multiple languages would be awesome.
 * Yaris678 (talk) 06:41, 13 April 2012 (UTC)
 * I don't think that meiskam's changes included any "bug fixes". From my view he encoded two things: (1) localization, and (2) persistent settings. The latter was integrated into the 2012-04-11 release -- and if localization becomes a reality then I will certainly make the effort to integrate the former. Regardless, meiskam's work is appreciated! Thanks, West.andrew.g (talk) 16:56, 13 April 2012 (UTC)

Risks of using Stiki
Perhaps a bit more information is needed about the risks of using stiki. Example: does it crash your computer.--Deathlaser (talk) 14:47, 12 April 2012 (UTC)
 * Message also posted at Reference desk/Computing
 * It has never crashed my computer. It runs through a Java virtual machine so I would have thought that crashing would be pretty unlikely.
 * That said, how much experience do you have of Wikipedia? I wouldn't recommend using STiki if you haven't already been around the block a few times.
 * Yaris678 (talk) 15:28, 12 April 2012 (UTC)
 * Does that mean I am restricted from using stiki.--Deathlaser (talk) 16:18, 12 April 2012 (UTC)
 * No. There is currently nothing in the software to stop you from using it.  However, if you end up doing something wrong with it then the fact that you are new to Wikipedia is not an acceptable explanation.  There is the chance that you will be blocked from using it, or even blocked from using Wikipedia.
 * Yaris678 (talk)
 * I myself have experienced no technical problems when using STiki. However, as Yaris says, combing the recent changes with a tool such as this is a delicate operation, and should be exercised with caution, especially if you're new to Wikipedia. Orphan Wiki (talk) 20:32, 12 April 2012 (UTC)
 * I have the highest AGF percentage on the leader board, is that a good thing or a bad thing.--Deathlaser (talk) 14:33, 14 April 2012 (UTC)
 * See my response in the explicit section you started below. Thanks, West.andrew.g (talk) 15:55, 14 April 2012 (UTC)

AGF percentage
I have had a look at the leader board and noticed I have the highest AGF%, is that a bad thing? Is high numbers of AGF a negative thing? By the way, Stiki is 10X faster than twinkle. Thank you for this amazing tool.--Deathlaser (talk) 14:37, 14 April 2012 (UTC)


 * Hi Deathlaser, and welcome to STiki. Fundamentally, there is no issue having the highest AGF percentage, I guess this would imply you are someone you "assumes good faith" alot. What's most important is that the AGF edits you classify as such actually fit the criteria. You certainly shouldn't be reverting good edits; and if you are AGF'ing malicious vandals then they aren't getting the warning templates that will lead to them getting blocked. As long as your are accurately classifying and don't have a case of edit-count-itis, proceed as usual and don't concern yourself too much with the leader-board. Thanks, West.andrew.g (talk) 16:03, 14 April 2012 (UTC)


 * Hi Deathlaser,
 * I'd say you probably are tending towards assuming good faith a bit too much. For example this revert didn't need to be "good faith" revert.  You can just call it vandalism.
 * If you look at WP:AGF it says "This guideline does not require that editors continue to assume good faith in the presence of obvious evidence to the contrary".
 * Don't worry about the reverts you've already done - the important thing is that the vandalism was reverted. But do think about if you have "obvious evidence to the contrary" next time.
 * Yaris678 (talk) 23:30, 15 April 2012 (UTC)

Advert Bot
I want to make a bot to notify thousands of active users who have Rollback to use Stiki.--Deathlaser (talk) 16:59, 16 April 2012 (UTC)


 * I'll get into the details when I have a little more time to post, but I urge you NOT to do this. Foremost, you'd need to get BAG approval. Second, its spammy behavior and probably would not be well received by many. Thanks, West.andrew.g (talk) 18:13, 16 April 2012 (UTC)


 * What is "bag approval".--Deathlaser (talk) 19:09, 16 April 2012 (UTC)


 * Please see WP:BAG. Thanks for the thought! Allens (talk &#124; contribs) 19:13, 16 April 2012 (UTC)

Talk page redirects
Does STiki really want to follow talk page redirects? (T#002) It seems quite an obvious way for a vandal to deflect warning from his/her talk page - make it a redirect to somewhere else. Someone could even use it as a way to annoy another user. If we do implement this feature it would have to be an option for the user - alert the STiki user to the presence of the redirect and make them work out what the best action is. Yaris678 (talk) 06:36, 13 April 2012 (UTC)


 * Valid point; However how many vandals even know about redirection functionality? If a user wants to avoid warnings and annoy other users, this is just the tip of an iceberg. But yes, this is probably rare enough that some dialogue could be displayed in the odd case it happens. Thanks, West.andrew.g (talk) 17:05, 13 April 2012 (UTC)

WikiTrust queue
Question: Is the WikiTrust queue built: Doing the latter might help its otherwise-abysmal accuracy, if it's not already being done. Allens (talk &#124; contribs) 02:50, 15 April 2012 (UTC)
 * Purely off of the WikiTrust reputations of those making edits; or
 * Also off of the level of estimated quality of the material being edited (as in, how enduring it has been previously)?


 * The WikiTrust queue is built using the "quality API" as described on WikiTrust's homepage. I am professional colleagues with the WikiTrust folks, so including their approach was a logical part of STiki's evolution. I must say, I am not too eager to open up WikiTrust and start fooling around with how it computes its values. I've recently used it in some related research (measuring author quality in code repositories), and between the OCaml language and its overall complexity, these are not trivial matters. However, I find it interesting to imagine why the WikiTrust queue has been *so bad*. Could this just be a function of the fact no one uses it? If no one used the STiki queue, for example, might all the false-positives just remain atop the queue and frustrate users? Thanks, West.andrew.g (talk) 05:37, 15 April 2012 (UTC)


 * Hrm... looking over the site in question reveals what was bothering me in the back of my mind: I had previously read over the lab report, and it confirms that they are indeed using the level of estimated quality. Oops! Sorry about that... This finding indicates that what may be more applicable is to add the minimal-NLP features (I'm guessing, among other things, "Bold text" and similar, profanity/obscenity, etc detection?) that the STiki queue uses in addition to the metadata.
 * Your question regarding false-positives remaining "atop the queue" brings up one further question that I've wondered about: How are the queues ordered? Most likely to least likely vandalism, or what? If it is most likely to least likely, then the false positives should not be atop the queue... Allens (talk &#124; contribs) 22:51, 15 April 2012 (UTC)


 * Queues are strictly ordered based on vandalism probabilities. They are priority queues (in a DB table) where the vandalism probability is the priority for insertion. Edits are dequeued if (a) a new edit is made on the same page (and that new edit will soon be enqueued), or (b) a classification is made in the GUI.
 * Also, if a queue is not used for a prolonged period, false-positives *will* start seeping to the top (lowering accuracy) -- as strange as it sounds. Basically, people are likely to eventually find the spam/vandalism (without STiki) and though STiki might have had these near the top of the queue, they get dequeued when someone else finds/handles them. When these "fall out" what's left behind are the false-positives near the top, since these pages are less likely to be edited in the same time-frame (because the change was good, and the article is in a good state). Of course, new edits are always streaming in, but the phenomenon exists. Every so often I should just wipe the queues clean for a fresh start, maybe. Thanks, West.andrew.g (talk) 23:39, 17 April 2012 (UTC)
 * Or perhaps wipe out everything older than x days that hasn't been checked out? Allens (talk &#124; contribs) 23:49, 17 April 2012 (UTC)
 * Andrew, I don't think lack of use alone can explain the poor performance of the WikiTrust queue. Lack of use would mean that false positives wouldn't be removed... but it would also mean that true positives wouldn't be removed... so the density of true positives shouldn't change.
 * I didn't read far enough ahead, but this is a clearer way of stating what I did immediately above -- the density of trust positives is changing in my assessment, because they are getting found by non-users of STiki. Thanks, West.andrew.g (talk) 23:42, 17 April 2012 (UTC)


 * My take on WikiTrust's performance is that it measures accumulated trust but has no measure of distrust. To put it another way, it has no way of distinguishing a vandal from a new user.  I am not exactly sure why this is... obviously there are some IP addresses that are responsible for a lot of vandalism and you'd think it would be able to determine that these users had a good portion of their contributions removed... but maybe it was not designed that way.  You'll have to ask your colleagues!
 * Yaris678 (talk) 23:06, 15 April 2012 (UTC)
 * First, we need to distinguish the core WikiTrust algorithm (described in a WWW'2007 paper) versus the anti-vandalism machine-learning classifier which is also "WikiTrust" (described in a PAN-CLEF paper from 2010 -- not 2011). The former has a very weak notion of distrust, as it is possible for have users to have a reputation below the initial value (barely). However, it is the latter that really accounts for this case. The ML classifier wraps the WikiTrust algorithm score with other features like "is IP?", "comment length", etc. that helps in these cases. Thanks, West.andrew.g (talk) 23:56, 17 April 2012 (UTC)


 * I like Alans's idea of adding in NLP feature’s to the WikiTrust queue... but rather than making it minimal... how about using the CBNG score? Calculating the probability of vandalism as a function of CBNG and WikiTrust scores could be the best of both worlds!  I think it would be fairly straight-forward to create such a function.  If I had the data, I would do a first pass of kernel smoothing and then work out the values close to 0 and 1 using log odds and a second pass of kernel smoothing.
 * Yaris678 (talk) 23:48, 15 April 2012 (UTC)
 * The CBNG, meta, and WikiTrust queues all have fundamentally different approaches to edit scoring. This is a good thing, and it is my intention to leave these 3 queues intact. The logic here is correct though -- we should be combining these different approaches. This is the niche the "meta queue" (a greyed out option) is intended to solve. We've already done this in an academic/offline fashion (CICLing 2011 paper focus on page 8). We didn't use the CBNG approach (they aren't the best at using the same datasets as the academics), but did encode lots of language features in that theme. The issue is bringing this model online (and determining what features are actually worth the time they take to calculate). So rather than adding "a little of this and a little of that", the approach should really be "add it all together" (and find the time to do this). Thanks, West.andrew.g (talk) 00:07, 18 April 2012 (UTC)


 * In terms of the WikiTrust queue... I'm seeing entries for Cluebot-NG not being considered trusted. Allens (talk &#124; contribs) 11:53, 16 April 2012 (UTC)
 * Ditto for a large number of other bots, plus User:Materialscientist (an admin active in reverting vandalism). It looks like my suggestion above is not quite what's needed - having both a low (no higher than an IP address at the start) user reputation and changing something with a high trust value (not just one or the other) should be required, at least if there's not other information such as a high CBNG score. As it is, it looks like that if something hasn't gotten changed previously (e.g., the interwiki links tend to remain stable), even a user who should have a high reputation can get marked as a possible vandal. Allens (talk &#124; contribs) 12:43, 16 April 2012 (UTC)
 * Hmmm... well... I kinda can see why CBNG and Materialscientist could have low scores... but I thought it was supposed to be designed so that if you are only ever reverted by users with low reputation your own reputation doesn’t suffer too much. That’s what it says on point 3 on the FAQ.  It sounds like this feature isn’t working properly.  Maybe it needs to start by assuming that admins and approved bots are not vandals.  Or maybe there is some bug in the way WikiTrust’s numbers are being fed into STiki.
 * I think it would help if Andrew had a chat with the WikiTrust folks.
 * Allens, have you got diffs for the edits by CBNG and Materialscientist? That could help in debugging.
 * Yaris678 (talk) 15:39, 16 April 2012 (UTC)
 * The description above is pretty accurate of how I understand WikiTrust to work. I can see how pure anti-vandals have a hard time accruing reputation: they rarely/never author novel content (they just restore content someone else wrote) and they are probably involved in plenty of edit warring (and every CBNG false-positive will be a serious mark against its reputation). Some factors should remedy this, but maybe its not enough. Thanks, West.andrew.g (talk) 00:18, 18 April 2012 (UTC)


 * No, but they can be located in the queue that I've marked as innocent. I'll post any further such diffs I see. Allens (talk &#124; contribs) 16:09, 16 April 2012 (UTC)
 * Oh, BTW: I agree with you that the trust scores should really be initialized in some respect - assuming they're from -1 to 1, with 0 being neutral and 1 being best, maybe something like 0.1 for approved bots (and users with "reviewer" if the current pending changes RFC goes through), 0.2 for admins, 0.3 for bureaucrats, 0.4 for oversight. These could also be used for minimum reputations as well as starting ones. Allens (talk &#124; contribs) 16:25, 16 April 2012 (UTC)
 * The WikiTrust algorithm does have initial values. I think the minimum reputation is 0, and new users have initial reputation 0.1 (and a maximum around 20,000). Once again, I stress this is the ALGORITHM. When you call the API you are getting a probability from an ADTree which includes many more features than just that value. Thanks, West.andrew.g (talk) 00:22, 18 April 2012 (UTC)
 * An example bot edit: 487457931 for Ammonium hexafluorophosphate - that's an inter-wiki link addition in the WikiTrust queue. Allens (talk &#124; contribs) 16:31, 16 April 2012 (UTC)
 * An example CBNG edit: 487696610 for Mohanlal Chaturbhuj Kumhar Allens (talk &#124; contribs) 17:37, 16 April 2012 (UTC)
 * This looks wrong to me:
 * {|cellspacing = "0px", width = "95%"


 * - valign="top"
 * Article
 * Edit
 * User
 * WikiTrust "likelihood"
 * 1.
 * Nissan 370Z
 * 487586327
 * Yaris678
 * 0.219942755011745
 * 2.
 * Mohanlal Chaturbhuj Kumhar
 * 487696610
 * ClueBot NG
 * 0.219942755011745
 * }
 * Two different articles. Two different users.  Identical likelihood!
 * Have I called the API wrongly?
 * Is this some kind of default value, meaning that WikiTrust hasn't calculated the correct value?
 * Whatever the reason, I think this could be related to the poor performance of the WikiTrust queue.
 * 18:10, 16 April 2012 (UTC)
 * It should not be unusual to see "identical" WikiTrust scores. WikiTrust uses an alternating decision tree as its scoring model. Given the finite number of paths through such a tree, there is a finite number of probabilities that WikiTrust computes. I'll address the comments in the rest of this post at a later time. Thanks, West.andrew.g (talk) 18:17, 16 April 2012 (UTC)
 * OK... so the use of ADTree explains why it is not totally unbelievable that the scores would be identical... but given that we are comparing an anti-vandalism bot to a user it still seems very unlikely. Why would the ADTree put two very different users in the same category?  Yaris678 (talk) 18:43, 16 April 2012 (UTC)
 * 18:10, 16 April 2012 (UTC)
 * It should not be unusual to see "identical" WikiTrust scores. WikiTrust uses an alternating decision tree as its scoring model. Given the finite number of paths through such a tree, there is a finite number of probabilities that WikiTrust computes. I'll address the comments in the rest of this post at a later time. Thanks, West.andrew.g (talk) 18:17, 16 April 2012 (UTC)
 * OK... so the use of ADTree explains why it is not totally unbelievable that the scores would be identical... but given that we are comparing an anti-vandalism bot to a user it still seems very unlikely. Why would the ADTree put two very different users in the same category?  Yaris678 (talk) 18:43, 16 April 2012 (UTC)


 * Backing off the indentation here. I've just gone through and placed my own feedback in-line. My main themes are: (1) There is a difference between the core WikiTrust algorithm and the machine-learning model which has been wrapped around it; this can explain much confusion. (2) We shouldn't be improving individual queues, we should be combining all available logic for the (super) (meta) queue. Take a look. Thanks, West.andrew.g (talk) 00:25, 18 April 2012 (UTC)
 * I agree that the meta queue is of interest, although I worry that a number of the features being used are actually the same (e.g., length of comment field), so it may be worth it to look at improvements of the differing features. In regard to the WikiTrust queue, wouldn't the false positives be pretty old in it if your theory is correct? I'm seeing a number of false positives that are no more than a day or two old. Allens (talk &#124; contribs) 01:30, 18 April 2012 (UTC)
 * Yes, some of the features are the same. This is not a consideration when the meta-queue is built at the feature level (rather than just using the 3 output scores). It would be an issue if only using the 3 scores, but would still probably have a measurable benefit.
 * I don't think the false-positives have to be "old" in an absolute sense. I was just stating the predicted life of a vandalism edit is far shorter than that of a constructive one. You'll have an easier time finding day-old false-positives than day-old true-positives, I'd bet. Thanks, West.andrew.g (talk) 01:51, 18 April 2012 (UTC)
 * Some thoughts:
 * 1. Thank you for clarifying the situation with there being two different things that we call WikiTrust. Would it be fair to call these.
 * a. A WikiTrust score, which is the raw result from the WikiTrust algorithm.
 * b. A WikiTrust probability of vandalism, which is based on passing the score and other things through an alternating decision tree.
 * 2. Am I right to think that a is this and b is this?
 * 3. I agree that it makes sense to combine things at the right level. A combined queue should use a, rather than b.
 * 4. Are you familiar with regression decision trees? These give a different estimated linear relationship between variables, depending on the value of the independent variables... as opposed to ADTree, which gives a probability, depending on the value of the independent variables.  I think the ultimate answer to a combined queue might involve giving a different estimated non-linear relationship between vandal probability and WikiTrust score, CBNG score and a some function of spatio-temporal data, depending on the value of other bits of meta data (like whether or not IP and number of edits made).  However, I can imagine that will need a lot of human fiddling to make it work.  i.e. it couldn’t be automated to the extent of alternating decision tree or regression decision tree.
 * 5. Thank you for your explanation of how false positives can build up due to vandal edits remaining "top" for shorter periods. If this is the full explanation of why the WikiTrust queue is so bad then it does suggest that the WikiTrust probability of vandalism isn’t much good.  Other queues are frequently used to revert vandalism that is days old and occasionally months old.  "False positive concentration" appears to be less of an issue for them.  Of course, WikiTrust score could still be very useful as part of a combined approach.
 * 6. If the score can be between 0 and 20,000, and a new user has a score of 0.1, that does rather confirm my suspicion that the WikiTrust algorithm is unable to distinguish between a new user and a vandal. Obviously, 0.1≠0, but it would be nice to see more scope for accumulating distrust.
 * Yaris678 (talk) 09:42, 18 April 2012 (UTC)
 * I believe all of the above to be correct. Thanks for your suggestions. West.andrew.g (talk) 16:14, 18 April 2012 (UTC)

The "Wiki-diff" link
I like the fact that the STiki interface now shows the full diff to be reverted. However, I think the "Wiki-diff" link on the GUI needs to be similarly changed. I recently got this diff from the CBNG queue and clicked on "Wiki-diff" and got this diff. Yaris678 (talk) 19:29, 15 April 2012 (UTC)


 * Good find; I didn't propagate the "diff window"-> rollback change through all the metadata. This should be a straightforward fix. Filed as T#008. Thanks, West.andrew.g (talk) 00:34, 18 April 2012 (UTC)

External links from "cite web" not working
See 486025844 (for the Solar Lottery article) for an example - it shows up as if it should be clickable (although including the closing }} ), but nothing happens if one tries clicking on it. Allens (talk &#124; contribs) 23:19, 16 April 2012 (UTC)


 * Well, if my parser underlined the template closing marks: }}, then that is probably the reason nothing worked. What probably happened is that it tried to create a URL object using that URL (with trailing crap), and that constructor didn't like it (malformed) -- and the exception was caught and ignored (intentionally!) before anything serious broke. Parsing links out of a wiki-diff isn't as simple as you'd imagine; they can appear in plain-text, ref tags, and templates -- all with different rules over what "terminates" a link. I'll file this at T#009 and give it a look whenever I do the next batch-bug-fix. Thanks, West.andrew.g (talk) 00:46, 18 April 2012 (UTC)


 * In a possibly related problem, see the cite web|url= result at the bottom of 488109956 (for Pink (singer)). In this case, the display on the diff was malformed (multiple copies with a "> at the end), but not on the actual wikipedia page or wiki-diff. Allens (talk &#124; contribs) 03:24, 19 April 2012 (UTC)

Stiki behind a network proxy
Can stiki be used behind a network proxy/gateway? I tried using stiki and got a an error saying that there might be a problem with my network. Thanks. Staticd (talk) 08:51, 17 April 2012 (UTC)


 * See issue T#004 in the table atop this page. Simply being behind a proxy should not affect your use of the tool. More likely, either you or the proxy has a firewall installed which is prohibiting outbound traffic on port 3306 (MySQL). Could you test for this? Or maybe describe your network situation in a little more detail? Thanks, West.andrew.g (talk) 00:54, 18 April 2012 (UTC)

Link Spam queue: Whitelist? ...
I'm trying out the link spam queue, and finding that it often contains several instances of the same or very similar URLs. The most appropriate action (spam/good-faith revert/pass/innocent) for these is (in all the cases I've seen) always the same. This brings up two possibilities: I am also seeing some cases (e.g., 471989587 for the All the Devils Are Here article) in which there is actually no addition of a link, just its movement within the article; these cases need to be removed. As well as these to winnow down the queue, ordering the queue by likelihood of "improper edits" by the editor in question (judging by the STiki queue's and/or WikiTrust queue's evaluation) would also help. Allens (talk &#124; contribs) 12:57, 17 April 2012 (UTC)
 * Once someone has called "innocent" a given URL (or similar URL - hostname plus some portion of the rest) a couple of times, offer the option to put it in a personal whitelist (automatically classifying as innocent for the "link spam" queue) and/or report it to the server to be put into an overall whitelist once Andrew or someone else has gone over it manually. (For instance, there's a Philip K. Dick website that gets cited quite a bit in articles on his stories/books; for a less complex instance, see revision 483675050 (of the Vosburg article), with the addition of a (constant) reference used in various South African place articles.)
 * Once someone has hit "pass" for a given URL (or similar URL - hostname at a minimum?) a couple of times, offer the option to put it in a personal "greylist", of ones that the person can't evaluate, to be automatically considered "pass" for that person. (For instance, I usually can't evaluate a website that's in a language I don't understand. In this case, offering the option to personally greylist a country-level domain like ".ru" or ".kr" makes sense, albeit possibly in combination with the link-spam queue detecting the language of a site when it checks it out.)


 * Thanks for these suggestions. Relative the other queues, the link-spam one is far less mature, less-often used, and more experimental. Due to the lack of popularity, it hasn't gotten a lot of attention, and it is not a priority in my immediate future (I'd like to look towards the "meta" queue). That last bit should not be an issue. I compute both links "removed" and links "added" in a diff to handle the case of link-moves. What went on here warrants further investigation. Thanks, West.andrew.g (talk) 01:04, 18 April 2012 (UTC)

Odd characters in displayed link probably resulting in external link not being parsed
See 488373727 for Somatics. I suspect the trademark symbol in the displayed link is what's causing the problem. Allens (talk &#124; contribs) 08:20, 21 April 2012 (UTC)


 * I'm confused. I understand there is a trademark symbol, but how does this affect the link parse? The link does not parse at all? Or parses incorrectly? Thanks, West.andrew.g (talk) 13:48, 21 April 2012 (UTC)


 * The trademark symbol in the displayed link (using []) appears likely to be causing the problem, which is that the link isn't parsed at all - sorry for being unclear! I'm betting you used a regex for parsing external links and the portion to skip over the displayed text doesn't recognize the trademark symbol as something to skip over. Allens (talk &#124; contribs) 14:38, 21 April 2012 (UTC)

Upgrade to 1.20; Everything still working?
Hello all. In my casual browsing today I noticed that Mediawiki 1.20 was rolled out on en.wp. What brought this to my attention was the hideousness of the new diff format (but there are other places to discuss that fact; so let's not get into a debate here). More importantly, I wanted some confirmation that STiki survived the changes (it's suffered before, and I know Twinkle had some issues today). Hopefully it's just style-sheet changes for the diffs, so it will not affect the rather quirky way I parse them. I'm not sure the scope of all the changes, but do let me know if something seems amiss. Thanks, West.andrew.g (talk) 19:53, 23 April 2012 (UTC)
 * I haven't notices anything that I could put down to the new version of MediaWiki. I've noticed something else though... but I think that is unrelated so I'll stick it in a new section.  Yaris678 (talk) 20:44, 23 April 2012 (UTC)

Feature Request: Showing if wikilinks...
Quite frequently, such as with disambiguation pages and listings of "famous" alumni et al, whether something is vandalism or not can be best told by whether what has either been removed or added has a functional (non-red) wikilink. It would be nice if this were indicated in some way - perhaps with an underline for functional wikilinks? Allens (talk &#124; contribs) 03:19, 15 April 2012 (UTC)


 * Mmmm. Certainly a valid request. However, I need to think about this from an implementation perspective. I'll need to parse out all of the wikilinks (easy) and then probably ping the API to determine whether or not the page exists. I need to investigate how willing the API is to handle batch requests of this type -- it's certainly not unimaginable that there could be 10, 20, or more wikilinks in a single diff presentation. Bandwidth/latency is beginning to become a bit of a concern (consider for example, the delay after you log-in that it takes to display the next edit). Give me a bit to investigate. Thanks, West.andrew.g (talk) 05:27, 15 April 2012 (UTC)


 * Alternatively, could some use be made of whatever mechanism does the search-box autocompletion? Based on its speed, that doesn't seem to take up significant resources. Allens (talk &#124; contribs) 22:33, 15 April 2012 (UTC)


 * Filed as T#010 for future investigation. Thanks, West.andrew.g (talk) 00:07, 22 April 2012 (UTC)


 * The API responds perfectly well, for example: |Example2|WP:STiki|example3|Main_Page. I just tried putting 50 titles into this request, and it happily replied (50 is the limit).  If it returns a pageid the link exists, and if it returns a missing then it's a redlink. –meiskam (talk•contrib) 14:42, 23 April 2012 (UTC)


 * The concern is server load, bandwidth, and latency, not whether the API will do it. Perhaps it could be done only on hovering the mouse over a wikilink, with the results cached? A red underline could be used for a redlink, then, with a blue underline for a working wikilink, while no underline would indicate something hadn't been checked yet. It isn't generally necessary to show whether each and every wikilink in the diff is functional - usually just the ones that have changed are of interest, and sometimes not even all of those. Admittedly, it might actually be more efficient (since it's a single query) to do all the links at once, as long as there aren't too many of them (prioritize changed ones), but only do so on request for a specific diff, not for each and every diff. Allens (talk &#124; contribs) 16:01, 23 April 2012 (UTC)

One option would be to enable wikilinks in the diff browser. This would mean that a user could click on them and go to the relevent page in their normal web browser and hence check if it exists. This means that the existence of the page is only checked if the user cares. I guess this could be done in a similar way to how it is done with URLs. Potentially there would be parsing issues if the diff contains a change to a portion of the article name... which you could probably deal with by just disabling the feature in that small minority of cases. Wikilinks in the diff are included in wikEd and use them quite a bit. Yaris678 (talk) 16:13, 24 April 2012 (UTC)


 * I understand your suggestion, but:
 * What I am looking for is something placing less load on the servers and taking less time than, for instance, fetching the entire changed page, from which one could tell if at least any new links were live/dead (and similarly via the page history).
 * Cases in which the wikilink itself was changed are exactly the ones of most interest.
 * I can see your suggestion as an additional feature request, however. Allens (talk &#124; contribs) 21:36, 24 April 2012 (UTC)

Huggle
Someone mentioned wanting to use STiki queue with the Huggle application, I've thrown something together that will allow this. Application, and Source code. Make sure to tell it to sort by STiki. It's set up to pull the queue from the STiki IRC channel, which will leak your IP address if you use it. Also, because the only diffs coming in are from the IRC channel, diffs may get stale while sitting in the application, meaning they may not be the most recent. If someone is actually wants to use this long-term, I suggest the code be edited more to have it run better first. –meiskam (talk•contrib) 09:27, 20 April 2012 (UTC)
 * That sounds clever. I guess if someone were to improve it, they would have to make the Huggle GUI speak directly to the STiki server, as if it was a STiki GUI.  Have you mentioned this to the Huggle developers?  I notive they are currently working on version 3 of Huggle.  Yaris678 (talk) 11:29, 20 April 2012 (UTC)
 * The idea of an "anti-vandalism clearinghouse" which all anti-vandal tools cooperate with is a really-really good idea which I have been thinking about for some time. One of the biggest and dumbest wastes of energy is the notion of "implicit innocent." If someone finds a vandal edit, it gets undone by the first person to find it. If someone finds an innocent edit, nothing happens! Multiple Huggle users may investigate it, a STiki user might mark it "innocent", add some Twinkle folks and then the watchlisters -- all these people are just confirming what the first viewer knew! Anti-vandal tools should share information about their classifications and avoid duplicate work as much as possible. This is also a good way to quickly amass corpora, etc. Thanks, West.andrew.g (talk) 16:24, 20 April 2012 (UTC)

New version, Application, and Source code. This one works almost exactly like the original Huggle (it gets the edits from recent changes or the wikipedia irc, depending on your settings) and also has an option to sort by STiki. Outdated diffs will automatically remove themselves, and the STiki score is shown as a prefix in the queue. I think this version is good enough to use long term (I use it). –meiskam (talk•contrib) 06:13, 23 April 2012 (UTC)
 * Just in case - i compared the source code for the modified version with the mainline huggle client's sourcecode and the changes seem completely fine / harmless. Excirial ( Contact me, Contribs ) 18:39, 25 April 2012 (UTC)
 * What would be more helpful/interesting might be if actions in the Huggle interface could still fulfill the STiki feedback loop (i.e., marking edits as vandalism or innocent), so that STiki users don't end up with duplicate work. Of course, Huggle's lack of an "explicit innocent" classification might prove tricky. Thanks, West.andrew.g (talk) 20:05, 25 April 2012 (UTC)

Stiki and twinkle combined
Stiki is great at quickly finding vadalism, yet twinkle is great at giving warnings and using rollback. If only you could combine them so then when you make a revert, a new window opens in twinkle with a link to the page, Until then, I can't be using Stiki anymore.-- Deathlaser : Chat  15:19, 21 April 2012 (UTC)


 * But STiki automatically warns users in the case of blatant vandalism, if you have the checkbox ticked. For good faith edits, the message conveyed is contained more within the edit summary, rather than via a warning template. Are you saying you would desire a broader choice of warning templates from STiki? Because I find the warning system on Twinkle to be somewhat cumbersome, what with the swapping around with different windows and pages. Orphan Wiki (talk) 15:34, 21 April 2012 (UTC)
 * Finding it very difficult to tell between AGF and Vandalism.-- Deathlaser : Chat  15:57, 21 April 2012 (UTC)


 * I use both simultaneously. Stiki in one hand and twinkle in the other At the bottom of the Stiki Page there are several links clicking them will open in the browser where you can twinkle them, hope it helps --  Ð ℬig  XЯaɣ  20:16, 21 April 2012 (UTC)

Feature Request: Option to skip 3rd level warning for non-IP or autoconfirmed vandals
Hi. It has occurred to me that non-IP vandals (particularly if autoconfirmed) are potentially more dangerous, in that most of the tools at hand (CBNG, the STiki/metadata queue, etc) are less likely to identify vandalism from such, edit filters won't block most (perhaps any, depending on the current settings) actions by autoconfirmed users, semiprotection won't block such, etc. This is, of course, because vandalism is (much) less likely from non-IP/autoconfirmed users - but "less likely" does not translate into "less harmful when it does happen".

Given that STiki is less likely to spot vandalism from non-IP or autoconfirmed vandals, it appears advisable to try to move faster toward blocking these vandals when it does happen, in order to make up for this. This will require going faster to the 4th level warning (of "you will get blocked if you do this again"). It is undesirable to skip the 1st level warning, since it is more conciliatory if there has been a mistake (as is more likely to be the case if there have been no previous warnings); similarly, since the 1st level warning might be from a purely automated means such as CBNG, it would not be advisable to skip the 2nd level warning and go directly to the 3rd. Thus, an option to skip the 3rd level warning for either non-IP or autoconfirmed vandals appears desirable. (I suggest making it an option, not automatic, because some people may wish to be more gradual.) Allens (talk &#124; contribs) 20:54, 22 April 2012 (UTC)

STiki timeout
Is there a timeout on STiki? i.e. a point at which it starts to feed your edits to someone else? What is it? I was happily looking at this diff when I was beaten to the revert by Allens... who was also using STiki. I was taking my time because the previous version had a template formatting issue and I wanted to find a good version.

If we assume that the last thing I did on STiki before that was this edit, I must have been on it for 16 or 17 minutes when the Allens did his revert. (But I'm not sure, there may having been some passing of "innocenting" or in the interim.)

As a secondary issue... Allens didn't spot the formatting problem that I spotted so I still had to fix that.

Yaris678 (talk) 20:44, 23 April 2012 (UTC)


 * Whoops! Sorry about that. That's odd in terms of a supposed timeout - I've left STiki alone for hours and come back to find the same edits around with no noticeable increase in "already done" or other errors. Perhaps there's a persistent connection between STiki's frontend and the server, and that got cut off for some reason so the server assumed your computer had crashed/was disconnected? Allens (talk &#124; contribs) 21:11, 23 April 2012 (UTC)


 * No probs.
 * Yes... sounds like it could have been my Internet connection. It was mostly working at the time but I did have trouble at one point downloading a web page so I guess it might have just failed at the point when the STiki server asked my client if it was still there.
 * Out of interest, which queue were you using? I was using the STiki queue.  It would be interesting if you were using the CBNG queue and yet they had both decided that edit was the priority at that time.
 * Yaris678 (talk) 11:36, 24 April 2012 (UTC)


 * I most frequently use the CBNG queue, but I do switch around some. That's certainly another possibility for why it happened. Allens (talk &#124; contribs) 12:27, 24 April 2012 (UTC)

When one uses STiki, they place "reservations" on blocks of 10 edits (RIDs) at a time. Each edit in the block has a time-to-live of 15 minutes. After 15 minutes, if the edit is not classified in STiki, it is released to the general queue. That is part #1 of why Allens got an edit Yaris was originally in line to see.

When one gets a reservation, code immediately does all the data-fetching for all 10 edits. It caches this data so that revision advancement can be very quick, as there is a "local queue" of edits to be displayed. Now, every 1 minute (I think; it may be 30 seconds) we iterate through the local queue and ask "is this edit still most recent on the article?" If not, it is kicked out of the queue. Thus, Yaris' reservation not only expired, his display/revert of the edit had to come less than one minute after Allens succeeded at the revert.

Edits are released after a set period of time for a couple of reasons. First, on a normal "exit" action, STiki calls back to the server and erases the reservation on any edits that were not displayed. If a bug occurs or some irregular shutdown, this code is not run, and the edits are not freed (thus a time-to-live is necessary). Second, someone who leaves STiki open for hours without advancing shouldn't be hogging the most valuable parts of the queue (and as a hypothetical attack, someone could just open a ton of STiki clients to reserve out major portions of the queue).

There was a STiki user in the past who was very vigilant about correcting *everything* wrong with a page. Thus, he moved very slowly through his reservations, as he spent many minutes in the normal browser interface fixing issues (I remember the username/admin being "HTCMitchell" or something of the sort, but couldn't find it easily in my archives). I think in that case we just hard-coded a rule so his edits had a longer TTL. I'd be happy to do the same for any other regular STiki user. Thanks, West.andrew.g (talk) 03:55, 25 April 2012 (UTC)


 * Incidentally, I would not be keeping STiki open without using it if it remembered "pass" actions (for diffs still in the queue) between runs - my ability to read, say, Chinese is not going to significantly increase between two STiki sessions. Allens (talk &#124; contribs) 05:08, 25 April 2012 (UTC)


 * What do you mean? "Pass" actions are written persistently to the database and should persist for all eternity. The only time you'll see a passed edit "again", is if you restart STiki, the *first edit* you see may be a repeat. This is because when STiki starts, you have not logged in, therefore that first edit fetch is done under no specific username. Once you log in, everything resets and is re-queried, so the second edit onward will never be one that account has "passed" before. Thanks, West.andrew.g (talk) 14:36, 25 April 2012 (UTC)


 * Fascinating. (Wouldn't that be the first 10 edit fetches, not just the first one? Or am I confused?) Either this isn't working, or what I'm seeing are re-attempts by the same user to do something that someone else reverted. Allens (talk &#124; contribs) 14:41, 25 April 2012 (UTC)


 * At start-up, a reservation of 10 edits will be claimed and fetched under the "nobody" account. The first one will pop into your diff browser. I'll assume you log in at this point and classify that first edit. At this point, the "editor change" is caught. The "nobody" reservation is cleared, the local caching is cleared, and everything is re-queried and re-populated under user-specific settings (meaning you likely get the other 9 which were in the previous reservation). This is the delay after the first edit. Thus, only the first edit can be one you've previously ignored.
 * More than likely, your perceived "I'm seeing this again" cases, are really just cases of someone doing duplicate work on multiple articles. These edits tend to appear adjacent in the queue precisely because they have so many factors in common. Several times people have been insistent they were seeing the same RIDs, but it turned out to be tagging or redirection of very similar article names (i.e., "Boeing 747", "Boeing 747-700", "Boeing 747-700B", etc.). If you really feel like you've seen something twice, this is something that is trivial to audit. Thanks, West.andrew.g (talk) 20:27, 25 April 2012 (UTC)


 * I see - good design! I've also seen the cases of similar article names, but I'm pretty sure sometimes it's been the same article and the same edit, but - I would guess - by a different IP address or at a different time. Allens (talk &#124; contribs) 21:04, 25 April 2012 (UTC)


 * Andrew, Thanks for the info. I think I take longer than most people to check out an individual article but since this has only happened to me once I think I can tolerate that.  Changing settings for me would mean that I would be hogging the good stuff.
 * Andrew and Allens, I agree. If passes persisted between sessions that would be good.  If using "pass" properly then there is no point in seeing an edit a second time.  I suppose you could add an option to "purge passes" in case someone decides they want to go back and have a look at stuff they have passed already.  Yaris678 (talk) 07:31, 25 April 2012 (UTC)
 * See above. "Pass" actions are persistent across sessions. Thanks, West.andrew.g (talk) 14:38, 25 April 2012 (UTC)

Help required regarding Twinkle ... " the bottom of the STiki page there are several links ... "

 * ...but what if you are using STiki? Cannot use Twinkle at the same time, can you?
 * -- Gareth Griffith-Jones (talk) 11:55, 20 April 2012 (UTC)


 * I use both simultaneously. Stiki in right hand and twinkle in left At the bottom of the Stiki Page there are several links clicking them will open in the browser where you can twinkle them. --  Ð ℬig  XЯaɣ  12:12, 20 April 2012 (UTC)


 * That's good to know. Thank you very much!
 * Surely that must mean you are left-handed – I would have thought, hit the keys for STiki, and have the mouse for Twinkle. Isn't that correct?
 * -- Gareth Griffith-Jones (talk) 12:16, 20 April 2012 (UTC)


 *  THIS IS MY QUESTION: 

I cannot see the several links ... "At the bottom of the Stiki Page there are several links clicking them will open in the browser where you can Twinkle them" ... '''that User:DBigXray refers to above. Are you able to throw any light on this?''' -- Gareth Griffith-Jones (talk) 10:35, 26 April 2012 (UTC)
 * I'm afraid I don't know. As no one else has shown up to answer the question I would suggest asking User:DBigXray on their talk page, maybe they have the answer! Another good place to ask would be the STiki talk page. Someone there might be able to help you. I'm going to null the help tag for now, if you don't mind (if you do mind, you can replace it). Feel free to come along to my talk page any time with any questions. :) Ooh Bunnies! Leave a message 03:51, 27 April 2012 (UTC)
 * Posted here, following advice from OohBunnies! -- Gareth Griffith-Jones (talk) 05:43, 27 April 2012 (UTC)

Firstly, can I check that when you open up the STiki interface it looks like the screen shot to the right?

Secondly, all the blue, underlined words are links. For example, if you click on "Page-Hist" it opens the page history in your normal web browser; if you click on "User-Talk" it opens up the user talk page. This makes is easy to use Twinkle or to do any other thing that you would do in the normal wiki interface.

Does clicking on the blue, underlined words open anything in your browser?

Yaris678 (talk) 07:42, 27 April 2012 (UTC)


 * Excellent work, Yaris! I use Mozilla Firefox for Wikipedia.  Followed your instructions, and it opened up the user's Talk page in Internet Explorer.  Is that alright?  -- Gareth Griffith-Jones (talk) 08:34, 27 April 2012 (UTC)


 * I guess that means that Internet Explorer is set as your default browser. If you are happy to edit in Internet Explorer that's fine, but it should be fairly easy to change your default.
 * To set my version of Firefox as default, I open it up and go to Tools -> Options -> Advanced and click on the "Check Now" button under "System Defaults".
 * Yaris678 (talk) 09:03, 27 April 2012 (UTC)


 * Yes ,it is, but at Firefox Start I still get their message on start up, advising me that it is not my default, so I shall check it there. Again, thank you very much!  Cheers,  -- Gareth Griffith-Jones (talk) 09:17, 27 April 2012 (UTC)


 * So does that mean it's sorted now? When I get the message advising me that Firefox is not my default, it also gives me the option to make it the default.  Yaris678 (talk) 18:49, 27 April 2012 (UTC)


 * Yes. Just so.  And that is what I did.  So, yes, thanks to you, it is sorted.  Cheers!  -- Gareth Griffith-Jones (talk) 21:05, 27 April 2012 (UTC)


 * Cool. Happy to help.  Yaris678 (talk) 00:06, 28 April 2012 (UTC)

Congratulations to STiki on 100,000!!!
This afternoon STiki crossed the 100,000 revert threshold (vandalism + good-faith reverts combined). It was this good-faith revert by User:Excirial that did it. For perspective, revert 50,000 was this one, performed by User:Yaris678 on May 8, 2011. Here's to hoping that "250,000" isn't too far in the future! Thanks, West.andrew.g (talk) 19:18, 28 April 2012 (UTC)


 * Onwards and upwards! :) Orphan Wiki (talk) 09:30, 30 April 2012 (UTC)

CHANGELOG for 2012-04-11 STiki Release!! (and 2012-04-12)
Significant changes were rolled out in this version, causing the version number to bump up 2.0 -> 2.1. Many of these were front-end modifications to the GUI. These feature requests were a little easier to accommodate than I had expected, meaning: (a) STiki has a nice organizational framework or (b) I've screwed up and I am about to be bombarded with bug reports. The major changes are as follows:


 * Good-faith revert is now available as a fourth classification option. Good-faith reverts do not post any messages to the offending user's talk page. The comment left with good-faith reverts is visible -- and can be modified -- via the (newly!) tabbed comment panel (see a screenshot). The leader-board now also contains a good-faith ("AGF") column.
 * The notion of rollback is now fully integrated. Any revert action is now a "rollback" (including good-faith revert). For users that do not have the native rollback permission, we implement that functionality in software. To this end, the diff browser now displays all edits that will be undone (not just the most recent one). If more than one edit would be rolled-back, this is noted beneath the article title in purple font (see a screenshot).
 * Settings are now *persistent* between STiki sessions. This includes user-facing settings such as checkboxes, form fields, and window size/placement. Only *queue selection* is not stored; this remains server-set so users can be steered away from queues that are experiencing technical difficulty. Configuration is stored in XML format in the user's home directory. Special thanks to User:Meiskam for his assistance with this.
 * Backend: Logic added so CBNG queue will not enqueue edits made by bots.

As always, thank you for your continued use of the STiki tool. Please report any bugs you encounters. Shortly STiki will past the 100,000 revert mark, and this should be a great milestone in the software's history! Thanks, West.andrew.g (talk) 16:45, 11 April 2012 (UTC)


 * Piggybacking on this release, a few minor bug fixes and feature requests were fulfilled on new build released 2012-04-12. Notables fixes include:
 * The "activate hyperlinks" setting is now correctly read and applied from the user configuration XML. (T#007)
 * Functionality has been added so the diff-browser can be scrolled using the keyboard. The keys are PG_UP and PG_DOWN (slight scrolling) or UP_ARROW and DOWN_ARROW (more dramatic). The "classification" panel must have focus for this functionality to work (i.e., in a state where pressing "v" would fire a "vandalism" revert). (T#001)
 * Good-faith reverts will no longer be marked as "minor" (as vandalism/ spam edits are), per the policy of WP:Minor. Because native RB cannot be marked "minor", good-faith reverts must use "standard" editing calls to simulate rollback functionality ("software rollback"), incurring some processor/bandwidth overhead.
 * Thanks again, West.andrew.g (talk) 03:07, 12 April 2012 (UTC)

Two questions and a suggestion.
Before saying anything else i must say this first: While i didn't have the change to try STiki yet its description looks very promising. Especially the "Mark as innocent" is something i have been hoping to see for a long time - if Huggle has a drawback it would be that 100 people using it are about as effective as 5 due to checking the same thing over and over.

Two questions though:
 * If someone marks an edit as innocent, does this mean that this verdict is final (As in: The edit is never forwarded for a second check)? If so it might be sensible to offer an edit two times just in case the first editor misses something - Or in cases where someone is playing reverse vandal by marking everything they receive as innocent.
 * I would really suggest that T#005 (Qualification conditions) is implemented with the requirement for rollback. Other vandalism or fast-edit support tools require this functionality (Huggle and IGLOO require rollback, while Autowikibrowser maintains its own access list). I vaguely remember that was some talk about this being required for quality control sake. After all, incorrect or sloppy vandalism patrol is a rather large concern since it may drive away good faith editors. And just in case - is there a method to block someone from using STiki? For other tools it suffices to remove rollback, and Huggle will not work if its configuration page is protected.

And one suggestion: Excirial ( Contact me, Contribs ) 11:45, 25 April 2012 (UTC)
 * Would it be possible to implement key remapping? One of the greatest advantages of Huggle is that it is entiry keyboard-driven. (Q to revert and warn, R to revert, Space next edit, [] to navigate between edits). STiki seems to support is, but ALT+I, ALT+R and so on requires a lot more keyboard movement. Minor issue, i know. But thats why they call it a suggestion. :)
 * Hi Excirial,
 * I am inclined to agree with you about T#005.
 * Yes. Innocent is final (unless Andrew does a major change).  Of course, people can, by other means, still spot vandalism that slips through the STiki net.
 * There is some discussion that relates to your first two bullet points in the archived discussion Wikipedia talk:STiki/TalkArchive03.
 * Yaris678 (talk) 12:25, 25 April 2012 (UTC)
 * I'd like to point out that you can't really block any user from using an open-source program. Twinkle can be used by non-autoconfirmed users, AWB can be used by people not on the access list, and Huggle can be used by people without rollback.  I know all this because I've programmed a version of all mentioned programs that don't require any permissions.
 * As far as the keypresses, I don't think holding Alt is necessary (if that helps at all). –meiskam (talk•contrib) 12:35, 25 April 2012 (UTC)
 * Actually, it is possible to block an IP address, and possibly a logged-in user, from using STiki, since it is working of off a client-server model. Allens (talk &#124; contribs) 12:43, 25 April 2012 (UTC)
 * Thank you very much everyone - those are really helpful comments. I agree that you can never entirely block a user from using automated programs in case they are open source, as people can cobble their own custom rollback into those (Or even write a vandalbot from scratch). However, i assume that 99% of our vandals have no experience in coding something, and having to program something still takes more time then having to click "download". It is definitely a stopgap measure, but i assume it will work well enough for most part as i doubt that many people want to waste 30 minutes programming for vandalism.


 * As for the keys, your right indeed. On my laptop STiki utterly refuses to respond to key presses when i don't hold ALT, but my regular PC works just fine - i don't think i even want to try and figure what my laptop is doing. Either way i'm going to give STiki a whirl and see how it works; probably going to see if java is readable enough for me to change some keys. Or perhaps that Huggle version mentioned earlier is a nice idea, after i reviewed the code to make sure i'm not entering my password into a trojan :). Excirial ( Contact me, Contribs ) 17:39, 25 April 2012 (UTC)


 * Couple brief comments. First, STiki does have an internal blocking mechanism (only at my disposal, at this point) -- and a short-term block would solve any immediate and/or malicious issues. Second, there are keyboard hot-keys that do not require (ALT+ mnemonics). After a single edit has been classified with the mouse (thereby giving the button panel "focus"), the keys "V", "G", "P", and "I" will mark edits as "vandalism", "good-faith", "pass" and "innocent" respectively. While in the same mode, the PG_UP, PG_DOWN, UP_ARROW, and DOWN_ARROW keys will also scroll the diff browser. Thanks, West.andrew.g (talk)


 * Thanks for the info. I have been trying STiki for a short while, and it is definitely a useful tool for patrolling vandalism - i especially like that it serves "old" suspicious edits as most tools are dependent on the RC api and thus mostly limited to serving "recent" edits. This is very useful to mop up vandalism after times where patrol availability was low.


 * The keyboard layout is unusual though. While the used keys are entirely sensible name-wise, it requires an odd hand placement to use them. Left hand would use G and V (requiring a strange ring and middle finger placement) while the right would have to alternate between P and I to revert, and the arrow keys to navigate. Using only the keys to navigate is therefor somewhat counterintuitive - no criticism, only something i noticed while testing. Excirial ( Contact me, Contribs ) 20:46, 25 April 2012 (UTC)


 * Any suggestions on how it could be done better? A more ergonomic choice of keys would likely make the hotkey "letters" something that is very counter-intuitive with the classification labels. Thanks, West.andrew.g (talk) 22:11, 25 April 2012 (UTC)


 * Personally i am very satisfied (Or perhaps just very used to) the layout Huggle uses. Left hand ring finger rests on the Q (Revert and warn), index finder on the R (Revert), which can also switch to T (Template user only). The middle finder hangs around the E (Edit page), and the thumb rests on the space bar (Skip edit). If required the little finger can use the CTRL key together with Space to clear the entire revision queue. Right hand middle and index finger control the brackets, which allows navigation trough revisions. (While also being near the P which is used for Proposed Deletions)


 * The above is quite a natural hand placement, and allows easy access to all huggle's primary functions without using a mouse, which in turn means it is possible to keep ones eyes on the diffs at every time (Speeding up checking time and reducing eye strain). The buttons are not always to obvious name-wise, but i assume that most editors won't be looking at a keyboard to find a key and instead use a fixed hand placement where each finger matches some functionality. Excirial ( Contact me, Contribs ) 08:32, 26 April 2012 (UTC)

In my opinion the current location of Good faith revert(G) and vandal revert(V) are certainly not ergonomic for the left hand. I propose moving the good faith revert from G to A (A for wp:AGF) as i believe that the(A,V) pair is more comfortable than rather odd (G,V)  pair. any comments ? --  Ð ℬig  XЯaɣ  20:42, 27 April 2012 (UTC)


 * Feature T#005 (qualification requirements) was implemented in the 2012_05_22 release. A simple switch can force rollback entry requirements, but it will not be enabled without consensus; and I see no reason given the lack of abuse. I consider this matter closed. West.andrew.g (talk) 00:48, 22 May 2012 (UTC)

STiki documentation
I've got no massive issues with the content of this edit but the edit comment made me think: Yaris678 (talk) 18:46, 27 April 2012 (UTC)
 * If, in STiki, I go to any help, say "Help" -> "Help:Edit properties", then I get taken to the help on the offline review tool. This looks like the menu isn't referring to the right anchors in the help document.
 * 1) Minor issue, but why does it say "Help:" on each line of the help menu?
 * 2) I'm not sure what you mean by we should "just point users to the documentation at some point". Having info for users in a wiki means that it is a lot easier for people to add in stuff that they think their fellow editors will find helpful.
 * 3) While I agree that conciseness is good, we should also remember that the vast majority of the people looking at WP:STiki will be potential users.


 * My responses. 1: Works for me! Just tested on the latest version. Anyone else have issues? 2: No great reason; just to maintain the context of what was being browsed, I guess. 3 and 4: My point was simply that we need not re-write the documentation from scratch (I've written it once). I'd also like to keep the main WP:STiki page on the concise side to not overwhelm new users (I think we are pretty much at that sweet spot). That being said, for those who want more information, we should have a resource for them. I have no issue with creating a new wiki page and copying the STiki help-doc right into it for community editing. Thanks, West.andrew.g (talk) 19:16, 27 April 2012 (UTC)


 * 1: I get the same thing as Yaris678 (kept meaning to mention it at some point...) 3/4: Good thought on copying the help-doc onto a page for community editing. Allens (talk &#124; contribs) 19:25, 27 April 2012 (UTC)


 * Hmmm. How to sort out #1? It's not like anchor tags are all that complicated a matter. It works trivially on my system. The help-doc is simply an HTML document with standard anchor links. Where could things go wrong? Could this be system dependent? Thanks, West.andrew.g (talk) 19:44, 27 April 2012 (UTC)


 * ... and filed as T#011 for future reference. Thanks, West.andrew.g (talk) 19:47, 27 April 2012 (UTC)


 * It may be of interest to know that the internal links do fine - as in, going to the top or jumping to another section inside it via clicking on a blue link. It's the links from STiki's help menu that are the problem. Allens (talk &#124; contribs) 20:07, 27 April 2012 (UTC)
 * Internal links work fine for me too.
 * In case this helps with debugging, I am using Windows XP and Java 6 Update 29 (also known as Java version 6.0.290).
 * Yaris678 (talk) 22:53, 29 April 2012 (UTC)


 * These issues were resolved in the 2012_05_22 release. I now consider this matter closed (though it still isn't a bad idea to import and improve the full help document in wiki space). Thanks, West.andrew.g (talk) 00:49, 22 May 2012 (UTC)

