Restructure reboot

Awesome to hear! Okay it's been a week, let's get to work! I've closed a couple of tasks on the Phabricator and opened a couple more. I'm currently working on </nowiki>https://phabricator.wikimedia.org/T414766. I'm sorry I missed the conference, but my understanding is that we should proceed boldly. My opnion is that consensus on the "final" format will have a suspicious way of "emerging" once the pages are published, but that shouldn't stop us. I also believe that we should iterate and not "waterfall", to put it in software development management terms. Let's just get something out there quickly and get momentum going again. audiodude (talk) 04:37, 19 April 2026 (UTC)

@WhatamIdoing @Mathglot @Jenny8lee @SuperGrey @Aaron Liu (shamelessly pinging) audiodude (talk) 04:39, 19 April 2026 (UTC)

Okay, I know that people have strong feelings about LLMs/"AI"/Vibe coding, and that many in the community find it distasteful at best. I stand by the statement that I made last year in the WP1 repo, "That is, the tool should assist you not replace you." That said, I offerhttps://rspdemo.vibes.travisbriggs.com/, which is a web app interface for my existing RSP parsing tools. Running refresh downloads from live enwiki. Nothing ever gets posted. All browsing is done only on that server. audiodude (talk) 05:35, 19 April 2026 (UTC)

BTW I mostly did this for myself so I could iterate faster on the templates! audiodude (talk) 05:54, 19 April 2026 (UTC)

@Audiodude, what (if any) decisions/information do you still need, to get this done? WhatamIdoing (talk) 17:05, 19 April 2026 (UTC)

Thank you for that question! Currently the biggest blocker I see coming up is T414760, though Mathglot identified T414759 which is kind of a "sister issue". But I am not blocked at this moment. audiodude (talk) 19:34, 19 April 2026 (UTC)

I've added info at the Phab tasks, and you've responded. Most pressing is probably T414759, which is on me. The implementation notes suggest various solutions; Jenny and I talked about this at CredCon. The easiest (and ugliest) is implementation bullet 3, e.g. template WP:RSPSTATUS/sandbox gets copied to new template WP:RSPSTATUS2 for the next iteration, and ditto the other WP:RSP* templates. (A better implementation is described below.) I originally was concerned that having ugly naming like that would lock us in, as it would be hard to rename 500 pages later, but Oscar at CredCon pointed out that AWB could do that, in a rather boring, repetitive, but eminently doable operation. So, I am not *quite* as afraid of calling them WP:RSPxxxxxxx2 as I was before, although not crazy about it. Anyone concerned about this should comment at the task or here, and if no objection in a couple of days, I'll create the new template series, and you can use that in the next run.

There will be a new subtask added to coordinate with the WikiSignals group, as they are parsing the RSP table and are aware of the forthcoming change. I spoke with Barrett and Chris about keeping them in the loop, and possibly adding tokens (likely div's with classes, but t.b.d.) to facilitate parsing the new format. I'll look up their contact info, or maybe Jenny has it. Mathglot (talk) 17:23, 19 April 2026 (UTC) updated by Mathglot (talk) 01:40, 23 April 2026 (UTC)

Yes please put me in touch! There's actually an existential debate to be had around _the source of truth is wikitext scatter across 500 heterogeneously distributed pages_, versus the idea that it could be _a proper computer data structure somewhere (JSON or whatever)_. Just feels like a shame that we parse this table into a data structure, generate all these pages in 10 minutes whenever we want, and then immediately destroy the possiblity of ever doing it again. I know it's a crazy idea and a non-goal, what I'm discussing, but it's something I think about, in the realm of on-wiki/off-wiki/json/wikitect kind of things. Like, could we put RSP in Wikidata??Anyways /rant. audiodude (talk) 19:32, 19 April 2026 (UTC)

Yeah, one disturbing thing at CredCon was the rather over-reliance (imho) on the 500 (are we gonna end up making that a thing, and refer to it as The Five Hundred? (only half tongue-in-cheek) ) as some sort of source of Truth™. This is pretty much just what WhatamIdoing was afraid of, iirc; for example, in her militating against having a reliability status indicator on the index page (currently mocked up at WP:RSPINDEX, with others resulting from your runs); she was persuasive in that regard, at least for me. I remember thinking WaId would have some serious reservations about m:Cite Unseen. Anyway, while I totally get the 'beauty' (programmer-speak) of the data structure idea, my guess is that WaId would not mourn the demise of that possibility, since having it decentralized into 500 pages and making it harder to regenerate might be an advantage in her view, precisely because it makes it harder to batch process it. (But never fear; AI will undo that barrier anon, if not already. In addition, if you wanted to leave footprints to ease a recreation of the struct from the pages, such as hidden tokens at the unified CNET page manually merged from three CNET pages that provide clues how to map it back to three struct items, I wouldn't oppose.) Re WikiSignals: I'll get you the contact info. Mathglot (talk) 19:56, 19 April 2026 (UTC)

Yes, you're right: I have seen that kind of system abused. WhatamIdoing (talk) 20:18, 19 April 2026 (UTC)

Awesome, thanks for the perspectives and helping me crawl out of "programmer mind". i totally see the logic here and actually I agree. It's probably not a bad thing for there to be friction for creating or "canonizing" a RSP page. audiodude (talk) 21:03, 19 April 2026 (UTC)

Also interested to hear more theory and opinons about The Five Hundred. I really don't have a broad perspective on RSP in general, I'm late to the game and mostly just want to asist with the technical difficulties, as a technical editor. Do sources only have an RSP entry after an RFC? Is that a strict requirement? Why don't we just have a more free form way for editors to rank and discuss sources? (I'm sure this is something that was discussed at the conference and once again I'm sad I missed it). audiodude (talk) 21:06, 19 April 2026 (UTC)

The initial list was thrown together by a couple of editors based on their impressions and discussions they could find in the archives. There were no RFCs or other formal procedures; it was more like "I've answered that same question four times this year already, so let's make a FAQ already".

Entries tend to be treated like holy writ that has been carved in stone, which means they are rarely challenged and often applied either zealously ("I will track down and smite all of these!") or mindlessly ("Marked green. Brain off. Accepted"). All of this is bad for Wikipedia, so we want to minimize the number of entries.

Since then, we have imposed some requirements. This is partly because having someone decide that their life purpose is creating hundreds of new entries is Not Helpful, and also because long lists are difficult to navigate. We don't want editors to have to skim through a dozen sources that were never discussed, or only discussed hypothetically, to find the one or two that actually get discussed/disputed fairly often. WhatamIdoing (talk) 21:23, 19 April 2026 (UTC)

Thank you for the context! Yes I can see how RSP can both encourage lazy editing as well as gatekeeping. "Sorry no, appeal to authority, see WP:BLAHBLAH, you're not allowed 100% no exceptions, RSP says...." audiodude (talk) 21:37, 19 April 2026 (UTC)

Yup. It's so much more persuasive to say "RSP says..." than to say "WP:IMRIGHT and WP:YOUREWRONG WP:BECAUSEISAIDSO!" WhatamIdoing (talk) 21:45, 19 April 2026 (UTC)

Oh! Do tell. How have you seen it abused? That is very interesting to me. Jenny8lee (talk) 22:50, 19 April 2026 (UTC)

It mostly happens because people don't read the actual entries, instead just relying on the colour, or forget the "general" part in the legend of the table. "It's green so it's reliable" forgets that the point is that sources are only ever generally reliable, they can still be wrong in the specific context. Editors do the same with the other colours, being equally reductive.
The important point is that the list isn't authoritative, it's the consensus of editors that has weight. The red/yellow/green colouring tends to hide important details of the discussions it's based on.

So this is a valid rebuttal:

1. "You can't use The Times as a source as it's unreliable"
2. "The consensus is that it is a generally reliable source"Y

While this isn't:

1. "You can't use that Times articles as it's factually incorrect"
2. "The consensus is that it is a generally reliable source"N

-- LCU ActivelyDisinterested «@» °∆t° 10:45, 20 April 2026 (UTC)

With the scripts, the main problem is mindlessness on the part of the user: It's highlighted/has a 'bad' icon, so it's bad and should be removed. It's usually being done by a self-appointed Defender™ of the Wiki, whose favorite activity is reverting other people's contributions and whose purpose in using the script is to make those reversions faster and easier for themselves (NB: "faster and easier", not "more accurate"). Changing the documentation won't help, because the reason they're using these scripts is at odds with anything you would put in the documentation.

I think that some of these scripts could be useful for content contributors (e.g., to highlight that one Wikipedia reprinter whose name I keep forgetting). The problems don't arise from content contributors who are self-screening their sources. The problems arise from situations like this:

• A: This official media of the Chinese government is the perfect source for this Chinese government official's new title.
• B: Revert. My script says that's an unreliable source.
• A: An unreliable source? What do you mean?
• B: I mean that it's a biased propaganda source and generally unreliable.
• A: How could the Chinese government be "biased" about this Chinese government official's title?
• B: Huh? It doesn't matter what the sentence says. My script says the website is unreliable.
• A: But what source can you trust for Chinese government officials' names and titles, if not an official Chinese government publication?
• B: I have no idea what you're going on about. We can't use this source because MY SCRIPT SAYS IT'S BAD. Do you understand now?

WhatamIdoing (talk) 19:57, 20 April 2026 (UTC)

There's been an untick of editors wanting to have there favourite source be green listed, it's why I started the section above. I've also saw at a recent AfD that a source supported GNG because "/the latest script/ highlights it as green".
Of course we equally get those who defend the wiki by raging against what they see as "unjust" consensus, e.g. when it doesn't match their personal opinion. -- LCU ActivelyDisinterested «@» °∆t° 21:21, 20 April 2026 (UTC)

I agree with your points. There will always be people who misuse the scripts/icons, and therefore we don't want to encourage such misuse by putting more "nutshells" or broad coloring. For the RSP listing page, check out zh:WP:RSP, where the subpages are listed without any icons.
For Cite Unseen, I myself has been thinking about redesigning the icon, with a floating tooltip that presents more information (preferably, the whole RSP conclusion, and other metrics). I envision this could encourage script users delve deeper into the RSP rabbit hole rather than taking the face value of an icon. SuperGrey (talk) 21:50, 20 April 2026 (UTC)

Or maybe use an icon like a question mark or a brain, to show that some thought is needed?

One of the problems is that if the script pushes too hard against the user's preference, then they'll just switch to another script. You could end up with users thinking "Well, I used to like that Cite Unseen, but these days all it does is tell me to use my brain. I want a script that just tells me when I can revert; I don't want one that makes me think." WhatamIdoing (talk) 23:45, 20 April 2026 (UTC)

Yeah, to exchange convenience for educational value, the script might become less appealing and leading to downfall, unless -- an appeal to authority fallacy could be what lead those users to finally accept that "thinking is needed" -- by that I mean promoting Cite Unseen to the gadget list. Anyway, I would prefer the educational-value route. cc. @SuperHamster. SuperGrey (talk) 00:01, 21 April 2026 (UTC)

Indeed, this has always been a perennial concern. I have seen experienced editors remove valid instances of generally unreliable sources solely because they saw the scary icon. Our documentation, in bolded red letters, says that people should "do their homework" when evaluating sources but that's of course not what always happens in practice.

Cite Unseen's original purpose was actually moreso to denote the nature of sources (social media, government controlled, editable, press releases, etc.), but over time reliability ratings have also become a core feature as well (which I think is good!). The icons we use come from RSP but I'm all for considering alternatives that inspire more responsible use. ~SuperHamster Talk Contribs 00:36, 21 April 2026 (UTC)

I could start experimenting tooltip on zhRSP, as zhRSP already has subpages, thus easy to present RSP conclusions by transclusion. SuperGrey (talk) 00:07, 21 April 2026 (UTC)

WikiSignals contact info here. Also pinged you at Barrett's page. Mathglot (talk) 05:36, 21 April 2026 (UTC)

Facilitating parsing for partners

WikiSignals.org checking in. Thanks for all your work on the new RSP. Yes, we and a half-dozen other projects pull data from the RSPs in ~15 different langauge projects. And yes, we wish RSPs included machine-readable structured data — like media name -> domain -> status(es). Even better would be a regularly updated cache of the data from all the RSPs. I know that's not your mission, tho. But we can wish. Hearvox (talk) 14:05, 22 April 2026 (UTC)

• Adding subsection header and reindenting. Mathglot (talk) 18:53, 22 April 2026 (UTC)

@Hearvox and WebRunner95:, let's see what we can do to facilitate it. Portions of the generated subpages have unique classes, and/or we can look at adding span or div ids that by definition are unique. This would be quite easy now during the conversion prep phase. At a minimum, we can do it the way we would if we were going to be parsing the subpages ourselves as if we had no knowledge of a structured conversion, but if you want input into that process, this is the time.

P.S. Hearvox, not sure I got Chris's Wikipedia username right; can you lmk?

P.P.S., I labeled this subsection with the word 'partners' rather than 'WikiSignals', because it was clear at CredCon that numerous groups are parsing RSP. @SuperGrey and Jenny8lee:, do you recall who some of the others were? Mathglot (talk) 19:23, 22 April 2026 (UTC)

Hearvox wrote:

Looked at the demo: That'll make it even harder for us to parse the domain (and its RSP category. We, WikiSignals.org, use as a unique identifier ...

Can you expand on this? For example, see page Wikipedia:Reliable sources/Perennial sources/all/Dotdash Meredith, linked to from 'Dotdash Meredith' on the index page. Note that 'About.com' on the Index page also points to the Dotdash Meredith page, as do 'The Balance', 'Lifewire', and a few others. Is this the locus of the difficulty? Another way to implement that, would be to link 'About.com' on the Index page to Wikipedia:Reliable sources/Perennial sources/all/About.com instead, and then have that redirect to the Dotdash Meredith page. Would that work better for you? Mathglot (talk) 20:41, 22 April 2026 (UTC)

I think a better idea would be to integrate Cite Unseen-like rule params (m:Template:CULink) to the existing reliability infobox, and parse from that. If WikiSignals want to add their unique identifier to the infobox, they can also do so. SuperGrey (talk) 21:39, 22 April 2026 (UTC)

Your idea makes me think of a couple others:

• Use a hidden token similar to the hidden template name found near the end of the wikicode in every user warning template (example: {{uw-vandalism1}});
• Use a template which emits nil (like template {{void}}; could even be the void template itself, or call it {{rsp partner token}} or something) where the param value contains the needed WikiSignals token; e.g., {{void|1=wikisig-domain-or-other-info}}.
• Or maybe better: {{Rsp partner token|wikisig=domain-or-other-info}}, which would allow a natural evolution if partners other than WikiSignals get involved.

Mathglot (talk) 23:16, 22 April 2026 (UTC)

@SuperGrey, what kind of input do you need/want? I'm looking for an answer that sounds something like "I need to be able to extract the URL and the basic color-coded status from the RSP page" or "It would be convenient if the infobox had machine-readable information about whether the source has any of the following characteristics". WhatamIdoing (talk) 23:43, 22 April 2026 (UTC)

I think it's about what input WikiSignals wants. Or at least, that drives what SuperGrey would want, iiuc. Mathglot (talk) 23:50, 22 April 2026 (UTC)

Here is one example I envision would be useful to all partners, including WikiSignals and Cite Unseen, while not adding more scaffolds to the already complex Wikipedia:Reliable sources/Perennial sources/all/Dotdash Meredith. Basically,

{{Infobox source reliability
 | type           = website
 | status         = nc <!-- the general status that applies to all domains except specified below -->
 | shortcut       = WP:DOTDASHMEREDITH
 | last           = 2020
 | blacklisted    = yes
 | domain1        = about.com
 | domain2        = verywellfamily.com
 | domain3        = verywellhealth.com
 | domain3-author = Firstname Lastname <!-- like CULink's author param -->
 | domain3-status = gr <!-- specific status for indivual entries can also be listed -->
 | domain4        = verywellmind.com
 | domain4-date   = <= 2022 <!-- like CULink's date param -->
 | domain4-status = gu
 | domain5        = youtube.com
 | domain5-pub    = Very Well Mind <!-- like CULink's pub param -->
 | domain5-status = gr
 | rfc        	  = 
 | rfcdate        = 
 | wikisig_uid    = <!-- wikisig's token here; they can use a bot to auto-assign a token for new entries -->
 | anotherpartner = <!-- any other partner's token can also be added individually -->
}}

With this, any partners can directly pull from centralized {{Infobox source reliability}} rather than each adding their own templates and scaffolds. SuperGrey (talk) 00:55, 23 April 2026 (UTC)

I see what you mean. Fine by me, if it works for them. Let's wait on WebRunner95 (if I got the name right). Mathglot (talk) 01:46, 23 April 2026 (UTC)

@Mathglot re: your question above: I meant multiple pages are harder to parse than a singular page, but that's easily overcome. The {{Infobox source reliability}} looks wonderful. Exactly what we, and the at-least 6 other projects that machine-harvest RSPs, need. I'll catch up on this Talk page soon. For now, just wanted to add my support for the Infobox direction. (I also bugged Chris about commenting here: He uses WebRunner95 for Meta-Wiki, but, I think, @Localmotion127 for Wikipedia.) Hearvox (talk) 20:33, 27 April 2026 (UTC)

• Added {{tracked}} box above, linking Feature request T424586. Mathglot (talk) 21:48, 27 April 2026 (UTC)

Hearvox, thanks for the updated username; I've linked both of you at the new phab ticket. You mentioned six other projects that machine-harvest RSPs; can you list them? Mathglot (talk) 21:51, 27 April 2026 (UTC)

Partners (reference-reliability tools that pull RSP data):

• Cite Unseen
• Citatiion Watchlist
• Unreliable/Predatory Source Detector
• Veri.FYI
• Internet Archive Reference Explorer
• Iffy.news
• WikiSignals

—Hearvox (talk) 12:05, 29 April 2026 (UTC)

Thanks; phab:T424586 updated. Mathglot (talk) 15:43, 29 April 2026 (UTC)

RSP template refactoring

As mentioned above @17:23, 19 April, an important next step is to export the RSP template sandboxes currently being used in practice runs to a non-sandbox location, while not disturbing the live version currently supporting the RSP table. It's been three days with no comment, so I am going to move forward with this. I came up with a better solution than suffixing existing names with 2; details at T414759.

Once they are in and tested, I will update a couple of the mocked up source subpages using the new style templates, as a model output format for the next run. Mathglot (talk) 01:04, 23 April 2026 (UTC)