Wikipedia:Bots/Requests for approval/Lightbot 5


 * The following discussion is an archived debate. Please do not modify it. To request review of this BRFA, please start a new section at WT:BRFA. The result of the discussion was Symbol keep vote.svg Approved.

Lightbot 5
Operator:

Automatic or Manually assisted: Automatic supervised

Programming language(s): AWB, monobook, vector, manual

Source code available: Source code for monobook or vector are available. Source code for AWB will vary but versions are often also kept as user pages.

Function overview: Janitorial edits to units

Links to relevant discussions (where appropriate): This request duplicates the 'units of measure' section of Bots/Requests for approval/Lightbot 3. That BRFA was very similar to the two previous approvals: Bots/Requests for approval/Lightbot and Bots/Requests for approval/Lightbot 2.

Edit period(s): Multiple runs. Often by batch based on preprocessed list of selected target articles.

Estimated number of pages affected: Individual runs of tens, or hundreds, or thousands.

Exclusion compliant (Y/N): Yes, will comply with 'nobots'

Already has a bot flag (Y/N): No

Function details: Edits will add conversions to the following metric or non-metric units: foot, mile, mm, cm, m, km, plus their squares and cubes.

Discussion

 * I suppose it doesn't particularly matter if this appears under "Current requests for approval" or "Requests to add a task to an already-approved bot"; but for the record, this bot is presently flagless and blocked indefinitely to enforce Requests for arbitration/Date delinking. Lightmouse has an amendment before the Arbitration Committee, and the committee has indicated that any amendment is contingent on approval being granted by BAG, so the block and prevailing remedies are not necessarily hurdles with respect to bot approval. See related discussion at Bots/Requests for approval/Lightbot 4. – xeno talk  19:03, 16 August 2010 (UTC)
 * How will the bot know that it should not modify units which appear within quotations, since there is no rigorous way to identify quotations automatically?


 * In the early days of automation, this was a problem for everyone. However, AWB now has the very efficient 'HideMore' method for avoiding template, image, and quotes. Where Lightbot was updating templates, quotes weren't an issue so it had the option of running to the full extent of automation. For the addition of conversions it will be run with human supervision. Lightmouse (talk) 08:54, 17 August 2010 (UTC)


 * I cannot believe any software on the planet can automatically detect quotes when, as in Wikipedia, there is no requirement that the quotes be marked up with any particular tags. I am not just concerned about adding conversions, I am concerned with making any change whatsoever to units within quotes. I think you owe us an exact explanation, in plain language, understandable by those who do not write bots, of what kind of fully automatic changes will be made to units. Jc3s5h (talk) 15:14, 17 August 2010 (UTC)


 * Apparently it also looks for tagged quotes and double quotation marks (according to ["mask ... text between two quotation characters"]). A human will still need to detect any remaining quotes in single quotation characters. All conversions will be made with a human watching. There won't be 'fully automatic' changes. Lightmouse (talk) 15:34, 17 August 2010 (UTC)


 * The heading of this request states "Automatic or Manually assisted: Automatic". Any mention of semi-automatic edits contradicts the heading. I submit this request is malformed and must be repaired before approval can be considered.


 * The instructions say that 'Manually assisted' means "User must manually confirm every change". I take that to mean there is no option for auto-save even when the human is watching. It seems to have the effect of nullifying the application. It doesn't have an option for "User must watch changes just in case." If I've misunderstood, then please tell me what a manually-assisted bot can do that a normal editor can't. It might be a useful option. Lightmouse (talk) 16:23, 17 August 2010 (UTC)


 * I think everyone capable of judging whether certain changes carried out by bots are desirable is entitled to understand what proposed bots will do. If the structure of the Requests for approval page inhibits that understanding by not allowing accurate descriptions of bots, the structure should change. Could you state where the "instructions" you referred to are? Jc3s5h (talk) 19:22, 17 August 2010 (UTC)


 * Yes, communication needs to be clear. If this BRFA isn't clear, then we need to clarify. The instructions for how to fill in this form are still at the top of this page. It says "Manually Assisted: User must manually confirm every change"
 * I see that the automatic section actually says "Automatic: Specify whether supervised or unsupervised". On that basis I should have said "Automatic supervised". In previous incarnations of Lightbot, it said 'Automatic' because that was the worst case, the Lightbot 4 BRFA was simply a copy of the successful unit components of Lightbot 3 BRFA. All the discussion on Lightbot 4 focussed on the unit list and so I simply copied it again but reduced the unit scope massively. That explanation may not be acceptable to you but that is how it happened. Can you please tell me the difference between "Manually Assisted: User must manually confirm every change" and Not a bot? Lightmouse (talk) 20:12, 17 August 2010 (UTC)


 * Further, I stated my concern about any kind of edit to quotes, and you ignored that concern and just reiterated that conversions will be supervised. I interpret your unwillingness to assure us that the bot will not make any change to any quotation (that is, anything a well-educated human would recognize as a quotation, regardless of markup) as an acknowledgment that fully automatic changes will be made to some quotations. A specific example of such quotations are quotes that are indicated by indention, rather than the &lt;blockquote&gt; element, because of the strange quirks exhibited by the &lt;blockquote&gt; element. Jc3s5h (talk) 15:50, 17 August 2010 (UTC)


 * The phrase 'you ignored' suggests I'm being negative to you. If I misunderstood you, or you misunderstood me, I'm sorry. I took your point that no fully automatic system can detect a quote that has no indication other than indentation. I'm merely emphasising that a human is also in the loop and thus isn't fully automatic (a mode that's more suited to well-defined technical changes to templates). That may not be an answer that will lead to your support, but I said it with good intent. Lightmouse (talk) 16:23, 17 August 2010 (UTC)
 * The fact that there would be no fully automatic edits changes the complexion of the discussion entirely. My main concern with editor-approved changes is that the style and size of the window showing the editor the proposed changes might not provide enough context to know if the change is appropriate or not. Jc3s5h (talk) 19:16, 17 August 2010 (UTC)
 * OK. I think we're now focussing on a key issue. The three options: Automatic unsupervised (not being requested); Automatic supervised (I think this is the closest to what was requesting); and Manually assisted (I don't understand the difference between this and 'not a bot'). I think the two threads are merging now. Can we continue the debate at the bottom of the page? Lightmouse (talk) 20:17, 17 August 2010 (UTC)


 * How will the bot identify articles where a consensus exists that it would be overly repetitions to provide conversions for every measurement, and instead provides conversion factors in a footnote (or similar mechanism)? Jc3s5h (talk) 19:12, 16 August 2010 (UTC)
 * In all my time on Wikipedia, this issue has only cropped up a few times. One example related to maritime exclusion zones expressed in nautical miles. Another example related to weapons (old ship guns perhaps) expressed in inches. Those don't apply here because they aren't in the list of units. There is currently a debate going on about tables in US road junction lists. That doesn't apply here because they don't show the unit name in the table anyway so the code won't pick it up. Lightmouse (talk) 08:54, 17 August 2010 (UTC)


 * I interpret this to mean that the bot cannot tell if there is a consensus to limit the number of conversions, that Lightmouse has seen a few instances of this in the past, but by happenstance, those particular articles would not have been modified by the bot. I oppose bots that will ignore the consensus style of an article, even if it does not happen often. Jc3s5h (talk) 15:14, 17 August 2010 (UTC)
 * I don't know how a human can detect what consensus applies to an article. Lightmouse (talk) 16:32, 17 August 2010 (UTC)


 * (ec; not yet considering Jc3s5h's comment) Also, for the record, the bot is subject to a community ban, which may not necessarily be removed if Arbcom agrees to the BAG approval. (I still believe that to be the case, but I can't find any reference in the archives, so I'll strike my comment.)
 * That being said, this seems reasonable, provided
 * The list of changes to be made is published before or immediately after the any test runs, and any change in the code should be followed by a new test run.
 * It's made clear that only simple application of the units should be involved (e.g., no "foot pounds" or "pounds force", and "units" which may occur with a non-unit meaning should only be run in semi-automated mode)
 * An off switch should be provided for non-admins, in case the bot runs wild, as previous of his bots have done.
 * — Arthur Rubin (talk) 19:17, 16 August 2010 (UTC)
 * Could you link the community ban? – xeno talk 19:21, 16 August 2010 (UTC)
 * I can't find it in a fairly complex search of AN*, so I'll have to withdraw the comment. It won't be repeated unless I can find the link.  Perhaps it was during the time there was a separate Community Ban forum?  — Arthur Rubin  (talk) 20:26, 16 August 2010 (UTC)
 * WP:CSN? [is a subpage of WP:AN, so presumably would've been caught in a prefix search] – xeno talk 20:27, 16 August 2010 (UTC)


 * On second thought, by point 1 above, I mean the full list of transformations to be performed by the bot, in a form similar to the most detailed form presented in Bots/Requests for approval/Lightbot 4 (now withdrawn). — Arthur Rubin  (talk) 08:40, 17 August 2010 (UTC)


 * I request a copy of the AWB source code. My request is aimed primarily at learning more about AWB. Depending on how successful I am at understanding it, I might or might not make comments on the function of the bot based on source code. Jc3s5h (talk) 15:56, 17 August 2010 (UTC)


 * I haven't written the code yet. I'm glad I didn't because I've seen so many changes being discussed over the last month or so. And I suspect that you won't want to look at [] which I will be using to plagiarise. Remember that this doesn't just depend on code, several contributors appear to be unaware of target list processing, which is almost equally important. If you want to learn about AWB, you may wish to look at wp:awb. I still think it's easier to demonstrate maintain/convert units than to explain. Lightmouse (talk) 16:47, 17 August 2010 (UTC)

I've done a search of the Wikipedia database and identified 8 out of 3,385,487 articles that contain 'feet' or 'ft' between single quotes (about 2 per million). These articles can be modified or put on a whitelist. I hope that helps. Lightmouse (talk) 19:44, 18 August 2010 (UTC)

Query and suggestion
I have followed this page and the previous Lighbot 4 application. While the Arbitrators have said they’re willing to give the applicant another go at automation, by contrast, what I see here is a apparent presumption of guilt, an unwillingness to afford the flexibility of human input that is often central to good automation on WP—in this case, for dealing with the subtle and complex matters surrounding units of measurement. Such flexibility was given to the applicant until last year; it was largely successful, and enabled him to engage with the community and with individual users on many issues that would otherwise have remained otherwise dormant.

The application is for a time-limited, supervised trial. Lightmouse seems to have bent over backwards to accommodate concerns and to gain the trust of members, after the Arbitrators gave in-principle endorsement to the resumption of his work. The process has been going around in circles for many weeks. But the applicant is receiving a seemingly endless line of questioning in this BAG application that appears to seek ever more detail (such as comprehensive lists of units) before the code is even written or trials started; ironically, such questioning does not appear to be accompanied by any firm idea about the role of such detail in the application. While it is part of BAG’s role to probe applicants, this strategy is doesn't seem to be appropriate for the nature of the task that Lightmouse is applying to conduct as a trial. WP is riddled with fiddly little issues concerning the expression of units and conversions. Most of them go undiscussed, and remain in text in inconsistent or illogical forms. Many of them could and probably should be taken to WT:MOSNUM for discussion in the wider community. I suggest that Lightmouse is ideally placed, in running a trial, to identify some of these issues, using his considerable experience to refine both the social and technical aspects of unit editing. It is through such operation that issues might be discussed openly.

BAG should either say no or take the ball that Arbcom has passed it and approve a trial. It is not possible to assess the operation without a trial, so why not get on with it? If there is still concern, BAG might consider a shorter trial than the three months, with reportage of any issues at any time. But every indication is that the trial will be a valuable contribution to the project; I ask you to peruse, for example, a recent interaction about title consistency on LM’s talk page, to get a sense of his dedication to working through unforeseen and difficult issues with other editors. Tony  (talk)  08:51, 24 August 2010 (UTC)


 * Perhaps it is time for a trial; but the code and the list of transformations must be published (by Lightmouse) before the run; and reported errors must be corrected or consensus that they are not errors obtained before additional tests. — Arthur Rubin  (talk) 15:03, 24 August 2010 (UTC)


 * the Arbitrators gave in-principle endorsement to the resumption of his work - Not really. Kirill specifically stated "I would like to see a current statement from BAG indicating specifically which functions you will be performing" (emphasis mine). Most other arbitrators agreed with him. If anything, ArbCom has mandated thorough review and specific details before the request is approved and the restriction lifted. And if the code isn't finished, a trial would be premature for all involved. Mr.Z-man 15:36, 24 August 2010 (UTC)


 * Why is it so difficult getting somebody rehabilitated??? There seems to be so little trust and good faith. That, with the perennial drama of conflict, it's no wonder editors leave... Ohconfucius  ¡digame! 15:45, 24 August 2010 (UTC)


 * As I see it, there are many issues in Lightbot's previous incarnations, among which are:
 * Misunderstanding of his mandate. (Partially BAGs, fault, as they did approve the absurd "make changes in date formats".)
 * Bad coding, leading to the bot doing something he didn't intend.
 * And failure to recognize that, even when pointed out to him. (This may have have partially resulted from main point 3, below, which is not a problem, here.)
 * Failure to recognize that a consensus had not yet been obtained for his actions, in spite of BAG approval.
 * I don't see #3 as a problem here (except that he doesn't seem to note that quotes are not necessarily bounded by quotation marks.), but none of these require an assumption of bad faith, only of misunderstanding. "Rehabilitation" assumes that he did something wrong, and is willing to work correctly in the future.  These issues deal with mistakes, and, even in good faith, we need to establish clearly that he knows what he's doing.  — Arthur Rubin  (talk) 16:37, 24 August 2010 (UTC)

BAGAssistanceNeeded We've been discussing this for 6 weeks now. Units can be maintained/converted using supervised automation, it's been done successfully on thousands of small pieces of text throughout Wikipedia. If there isn't enough evidence already, then a trial run will provide more. If BAG has specific questions, I'd be happy to respond to them. The janatorial conversion and maintenance of units of measure is tedious by hand. It's an ideal task for automation using unremarkable and proven methods e.g. regex and target article list filtering. It would help greatly if BAG allow us to move forward to demonstration by example, i.e. the supervised trial stage. Lightmouse (talk) 16:55, 24 August 2010 (UTC)
 * Recused  MBisanz  talk 07:39, 27 August 2010 (UTC)
 * Propose a 50 edit trial. If there are resolvable problems, we can have another trial.  If there are unresolvable problems we can say "no". If there are no problems but people still have concerns we can have a 100 edit trial. Rich Farmbrough, 16:51, 27 August 2010 (UTC).


 * I'm a bit late to this discussion, but I oppose any automated addition of unit conversions to articles. A number of recent discussions have strongly indicated that there is no longer a consensus for the MOS guideline on units as it currently exists. Given the lack of consensus, we should certainly not be permitting anyone to make such edits by bot. Gatoclass (talk) 12:07, 30 August 2010 (UTC)
 * Sorry? Can you provide details? Which discussions, which consensus, and which aspecdts of the "MOS guideline on units". First I've heard of this. Tony   (talk)  14:22, 30 August 2010 (UTC)


 * Okay, here are a couple of links to previous discussions, there may have been more but I don't remember now where they occurred. Here's one discussion regarding precedence of units, and here's another concerning linked names. It seems to me at the least that the issues surrounding unit conversion are complex enough to make them unsuitable for bot automation. Gatoclass (talk) 04:33, 31 August 2010 (UTC)


 * Here's a couple more links to related discussions:.

There's debate in those four pages but as usual Wikipedia debates it's difficult to draw explicit conclusions. If conclusions have been documented somewhere, it might be useful to read them to see how they apply to this application. We've been discussing theory for weeks now without example edits. Last week I made a request for BAG input, so I hope it's ok to make another. Formal request for BAG input As Rich Farmbrough suggests, I propose a 50 edit trial. If there are resolvable problems, we can have another trial. If there are unresolvable problems we can say "no". If there are no problems but people still have concerns we can have a 100 edit trial. Lightmouse (talk) 10:43, 31 August 2010 (UTC)


 * I would like to see the source code available 48 hours before the trial takes place, together with a description of how selection of the article list or category will work in conjunction with the source code to minimize errors. Jc3s5h (talk) 11:17, 31 August 2010 (UTC)
 * A test batch of 50 represents 1/60,000th of the entire Wiki article population. Should anything go wrong, the risks are minimal. There is always the revert button. I am concerned that, with the above request, if the selection criteria are too narrow, the sample may be unrepresentative of the population of articles in mainspace. This would consequently risk greater potential disruption when a larger trial run is authorised because problems are not faced early on. Ohconfucius  ¡digame! 11:23, 31 August 2010 (UTC)


 * Ohconfucius, if the explanation is "the script is robust enough that it will work well on any article", that's fine. If the script has weaknesses that must be overcome by careful selection of the articles processed, that needs to be explained. Jc3s5h (talk) 12:07, 31 August 2010 (UTC)


 * Hey, pardon me, but it seems that you were the one implying a carefully selected list was needed. -- Ohconfucius ¡digame! 13:01, 31 August 2010 (UTC)
 * In the past the logic of Lightmouse's scripts would not be sophisticated enough to perform appropriate actions on any random article, but if the script were only allowed to process a carefully selected list of articles, then the weaknesses of the script could be averted. However, Lightmouse typically would explain how the script worked, but didn't explain his strategy in composing the list of articles to be processed, so the script looked like it would do bad things. So I am saying that if the script isn't robust enough to deal correctly with most random articles (and relying on his supervision of each edit to catch the ones that fall through the cracks) then the article selection strategy must be explained.


 * Given Lightmouse's customary way of working, I think the assumption must be the script WILL contain weaknesses that must be overcome by article selection, unless Lightmouse states otherwise. Jc3s5h (talk) 13:15, 31 August 2010 (UTC)
 * To err, is human. Yet you seem to be either setting a higher standard for Lightmouse, or you are assuming a lower level of competence. Either way, it's not 'charitable'. Also, how about some examples where things have gone wrong like you said, so that we are all clear what specifics you are referring to...? -- Ohconfucius ¡digame! 13:51, 31 August 2010 (UTC)
 * You haven't answered my question about what you think the risks are in giving the go-ahead on a batch of 50 articles. Pray tell... how would you select the 50??? Ohconfucius  ¡digame! 16:50, 31 August 2010 (UTC)


 * "Pray tell... how would you select the 50???" The script and the selection process go together; they must be designed in concert. Both must be made available so after an apparently successful trial, we will be better able to judge if it is really successful, or if it was just lucky and there are other articles around that would have failed. As for my personal preferences, I'd prefer to bug the Congress critters who take campaign contributions from companies who find it's cheaper to bribe provide political support for Congress critters than to modify their equipment to use SI. Then send the tapes to the Washington Post or 60 Minutes and wait for the blogosphere to demand Andy Rooney's birth certificate. Jc3s5h (talk) 17:56, 31 August 2010 (UTC)
 * I don't get this American political talk. Sure, the script is written with a job in mind. It's objective can be to change American spellings to British, or it can be to add templates to articles where there are 'naked' units of measure such as feet, miles, litres, hectares. But you want this palaver for a test of 50 articles??? shome mishtake shurely (sic). -- Ohconfucius  ¡digame! 02:24, 1 September 2010 (UTC)

We've been discussing this for six weeks now and I've been at pains to respond to detailed requests during this extended period; six weeks of talk seems quite enough for a 50-edit trial. I hope you'll forgive me if I now focus on responding to BAG. If BAG wants a trial to proceed, I'd be happy to develop and publish the code at their request. Lightmouse (talk) 12:01, 31 August 2010 (UTC)

A quick note
I should point out that the committee is expecting that a normal review of the proposal take place, but I see here scrutiny that is both unusual and difficult to justify given the relatively limited scope of testing at this stage of a bot request. In particular, some of the demands placed on Lightmouse appear to be unreasonable and designed to derail the process rather than borne out of a genuine concern for the technical accuracy of the proposed bot.

One of the basic principles on which Wikipedia operates is that of Assuming Good Faith; while Lightmouse was placed under a restriction because they had been (in the Committee's opinion) careless with automated editing in the past, they are now given an opportunity to resume their well intended contributions&mdash; and arbitrators will not look kindly on bad faith or attempts to sabotage the process. &mdash; Coren (talk) 23:19, 3 September 2010 (UTC)


 * I'm willing to assume good faith. The fact that Lightmouse failed to understand the scope (or, to be more precise, his bots were accused of exceeding both the stated scope and common sense)   of his previous bots suggests we should be more careful in describing the scope, so there would be less likely to be disagreement.  /4's proposed scope clearly exceeded common sense; but there could still be reasonable argument about whether an edit is in the proposed scope here.  — Arthur Rubin  (talk) 03:09, 4 September 2010 (UTC)

I think I will reiterate one point from above. The mere fact that a 50 article trial is approved and successful does not bind BAG to approve the BRFA (although BAG would normally do so). The question of which articles will be selected is not moot, for example I would expect astronomy articles to be skipped. However a test, successful or not does move things forward towards the eventual acceptance or refusal of this BRFA. Rich Farmbrough, 03:47, 7 September 2010 (UTC).

Mr.Z-man 04:08, 26 September 2010 (UTC)

Done. Lightmouse (talk) 10:14, 26 September 2010 (UTC)
 * I've looked through the edits and they seem to be good. Can you maybe comment about how these 50 articles were selected? Are they simply the first 50 from your master list for the final bot run? If so, could you give a comment about which articles are selected (or not selected) for this list? AKAF (talk) 08:59, 1 October 2010 (UTC)

There was more than one list. The list creation task involved trying to ensure examples that demonstrated the range of units. Each list that was created was then processed to eliminate some articles from the list (e.g. articles that define units of measure). I then ran AWB. As far as I recall, for each list, it was the first few items. I'm now making a formal request to either run the bot or to run a larger trial (e.g. 500 articles). Lightmouse (talk) 18:13, 13 October 2010 (UTC)


 * Seems a reasonable request given how the first trial went. . - Jarry1250 [Who? Discuss.] 20:43, 18 December 2010 (UTC)

Done. Lightmouse (talk) 11:39, 24 December 2010 (UTC)
 * I had a flick through a few and they seem good, but we can wait for feedback (ie. complaints), if any. - Jarry1250 [Who? Discuss.] 15:43, 24 December 2010 (UTC)
 * Okay, no complaints, and it is clearly in the operator's best interests to be careful with this one. - Jarry1250 [Who? Discuss.] 16:52, 1 January 2011 (UTC)
 * The above discussion is preserved as an archive of the debate. Please do not modify it. To request review of this BRFA, please start a new section at WT:BRFA.

Superceded by Lightbot 13. Withdrawn Lightmouse (talk) 17:10, 29 August 2011 (UTC)
 * Lightmouse (talk) 17:13, 29 August 2011 (UTC)