Post AxWCn5b0L0OObvUbYm by bnewbold@social.coop
(DIR) More posts by bnewbold@social.coop
(DIR) Post #AxU0MYksEGkhSflwRs by xgranade@wandering.shop
2025-08-24T01:00:07Z
0 likes, 0 repeats
Serious question/challenge to the ATproto folks: please write a guide as to how to get starred using ATproto if the domain bsky.app is blocked, and compare to writing the same guide for the case that mastodon.social is blocked.It's not totally contrived, given how hard fascist movements are trying to censor the internet.
(DIR) Post #AxU0MZj8cCxSTZy4Nk by xgranade@wandering.shop
2025-08-24T01:03:09Z
0 likes, 0 repeats
Like, honestly, how does someone --- without needing to be an expert in the protocol --- use ATproto without *also* using Bluesky PBLLC's infrastructure, signing Bluesky PBLLC's terms of service, and registering for an account with Bluesky PBLLC?
(DIR) Post #AxU0MaoqYT7PrZe8v2 by xgranade@wandering.shop
2025-08-24T01:07:34Z
0 likes, 0 repeats
If the claim that ATproto is a centralized protocol in practice is indeed "disinfo," what's the actual info?Because, from the outside, it all seems like a giant bait and switch where people make arguments for ATproto in the abstract, then conclude from that that we should most of us should go sign legally binding contracts with Bluesky PBLLC.
(DIR) Post #AxU0MbvGS5qXHlemYq by ricci@discuss.systems
2025-08-24T04:02:04Z
0 likes, 0 repeats
@xgranade I made the mistake of having a version of this conversation on bluesky this evening and the best argument anyone made for it being decentralized was one instance with HUNDREDS of users and another that might be up soon. Like maybe they will get there, it would be great if they do, but decentralized in practice is just so very transparently not where they are
(DIR) Post #AxU0MgOTpPap9XMAee by xgranade@wandering.shop
2025-08-24T01:21:21Z
0 likes, 0 repeats
I get the above might be unfair. That's not my intent at all, though — if I am wrong in the above, it's by trying to understand what I can about ATProto in practice as someone with a nontrivial amount of related skill and with a long lived history on the internet. If I'm missing something, I'd like to understand that better, if only to offer more fair and precise critiques of ATProto.
(DIR) Post #AxU0lx9mdnvANjX3nk by ricci@discuss.systems
2025-08-24T04:06:41Z
0 likes, 0 repeats
@SnoopJ @xgranade in fact the one second instance that was offered as evidence in that conversation was the same one offered up in this thread. Like it's great that it exists but my friends when one single instance is all that people have been able to point to for quite a long time now, sure it may be technically possible but it clearly isn't happening much FWIW I tried to set up my own PDS a while back and gave up. Maybe they've made it easier, I dunno
(DIR) Post #AxWCmxUKTwIdSuJA2q by bnewbold@social.coop
2025-08-24T01:41:25Z
0 likes, 0 repeats
@xgranade ATProto is designed to be broken up in to components which can be operated by different folks. This is a lot like the web/internet, but not like Fediverse, in that Fedi folks often expect to have one server combine "the whole stack".Some major ATproto components:- identity (DIDs)- hosting (PDS)- transfer (Relay)- indexing (AppView)- client app- moderation (Labeler)
(DIR) Post #AxWCmyvJ97TRurwDjc by bnewbold@social.coop
2025-08-24T01:44:05Z
0 likes, 0 repeats
@xgranade I'll give examples for each of this which aren't controlled by Bluesky PBC, and all work together. Nobody I know independently runs *all* these things in a *single* org.And i'm not going to claim that Bluesky isn't the dominant service provider for all of these for most people in the network. The intent is that there is at least real working proof-of-concept / running code for all these today (and for some time).
(DIR) Post #AxWCn05GpZ2NW3bgu0 by bnewbold@social.coop
2025-08-24T01:49:22Z
0 likes, 0 repeats
@xgranadeIdentity: I recommend most folks just use did:plc (which gives strong ownership and doesn't require domain registration), but did:web is supported and totally doesn't touch Bluesky PBC. Here is a (long, detailed, technical) guide:https://blog.smokesignal.events/posts/3lwopvsmtx22a-creating-a-did-method-web-identity-for-atprotocolhttps://handles.net/ is a cool service independent of Bluesky which shows how to give folks semi-custom domain handles at large scale (independent of PDS hosting)
(DIR) Post #AxWCn1MK5eGvTEap7Y by bnewbold@social.coop
2025-08-24T01:53:11Z
0 likes, 0 repeats
@xgranadeHosting (PDS): folks have been running independent PDS instances for more than a year now. It costs on the order of $5/month to run a full PDS with a small group of users. A recent community-scale effort has been Blacksky getting hundreds of folks to migrate.Most PDS instances require an invite code (otherwise spam bots sign up), similar to Fediverse.You can see a list of independent PDS instances here: https://blue.mackuba.eu/directory/pdsesthere are other dashboards around
(DIR) Post #AxWCn2MiLgBAajmeMy by bnewbold@social.coop
2025-08-24T01:55:17Z
0 likes, 0 repeats
@xgranade Relay: I have a blog post describing how to run an independent full-network relay here:https://whtwnd.com/bnewbold.net/3kwzl7tye6u2yA community member (phil) runs a handful of relays in different world regions, just for fun.Blacksky community has their own relay implementation and runs an instance.
(DIR) Post #AxWCn3FJ4hqdK3KFSi by bnewbold@social.coop
2025-08-24T01:58:01Z
0 likes, 0 repeats
@xgranade Indexing (AppView): focusing on bsky microblogging (ignoring things like blog posts, events, forums, etc), there are a couple different approaches to independent indexing, to serve things like reply threads.https://zeppelin.social/ is both a full-network bsky AppView, and a client app. It has been getting attention so might be slow right now.https://reddwarf.whey.party/ is another approach which builds on generic network indexes.
(DIR) Post #AxWCn3vqWXsLRzEDlg by bnewbold@social.coop
2025-08-24T02:00:19Z
0 likes, 0 repeats
@xgranadeClient Apps: these are usually pretty simple front-end apps which talk to AppView backend. But apps are still super important in atproto! They are ultimately where product features happen, and set defaults for which infra is used, and what moderation providers are required. as well as other policy decisions.The Bluesky client app is open source, and several folks run forks, like https://deer.social/https://deck.blue/ is a from-scratch independent client app
(DIR) Post #AxWCn4o5GtGEACbXJA by bnewbold@social.coop
2025-08-24T02:03:18Z
0 likes, 0 repeats
@xgranade finally moderation. which of course is the hardest and perhaps most important part of the whole system!The "Ozone" moderation service is open source and has self-hosting instructions:https://github.com/bluesky-social/ozone/There are dozens of instances focused on different aspects or "verticals" of moderation. A good example of a serious indie service is the Blacksky moderation team, which has more than 5 trained+compensated moderators (last I counted)
(DIR) Post #AxWCn5b0L0OObvUbYm by bnewbold@social.coop
2025-08-24T02:22:36Z
0 likes, 0 repeats
@xgranade that is a lot of components!if you have the perspective of wanting it to be easy to run the whole stack as a "single instance", this is a big pain (compared to, eg, the mastodon software distribution).from my perspective, it is *good* to have it broken up in to specialized components. they can be run by focused folks, recombined/reused externally. folks can build apps or contribute to moderation w/o having to host account data, for example.
(DIR) Post #AxWCn6agdfjThELrhg by xgranade@wandering.shop
2025-08-24T02:56:26Z
0 likes, 0 repeats
@bnewbold I appreciate the detail, thanks for the response. With all due respect, though, I think that kind of misses the question? I'm not unfamiliar with the protocol, and indeed have my xgranade.com domain pointed to Fed Bridgy; where I'm lost, though, is what *practical* advice I could give for someone who does not want to enter into a legal agreement with Bluesky PBLLC but who wants to use ATProto in the way its proponents describe. Ideally, that wouldn't take expertise in the protocol?
(DIR) Post #AxWCn7g2bFbr47regi by xgranade@wandering.shop
2025-08-24T02:59:55Z
0 likes, 0 repeats
@bnewbold Part of where I'm coming from is that it's quite simple on ActivityPub — just choose an instance that's not mastodon.social. That has been very widely critiqued as being too complicated, though. From that perspective, if ATProto is at least as decentralized as ActivityPub *in practice*, the answer to "how do I use ATProto without also using bsky.app" shouldn't be *more* complicated than "pick an instance."
(DIR) Post #AxWCn8OLwV3THYb2ky by bnewbold@social.coop
2025-08-24T03:27:54Z
0 likes, 0 repeats
@xgranadeThanks for the clear clarification!For an end user in a situation like that, ideally i'd send them to https://blacksky.community/, which is a client app, and has a "Create Account" button which will result in an account hosted by Blacksky, under their ToS (this would require an invite code). If they already had a Bluesky-hosted account, they could migrate it instead.
(DIR) Post #AxWCn98RB9uzaU9qaW by bnewbold@social.coop
2025-08-24T03:31:40Z
0 likes, 0 repeats
@xgranade the caveats there are the same with AP: how do you find these apps/hosts and get an invite? what incentivizes folks to run them? how independent are they really from upstream devs and other large players in the ecosystem? what are the economics of the moderation labor?
(DIR) Post #AxWCnAYLuIF3z9I3cW by xgranade@wandering.shop
2025-08-24T04:35:15Z
0 likes, 0 repeats
@bnewbold Right, so from the standpoint of a user who isn't deep into the weeds on social media protocols, the state of decentralization is that ActivityPub has lots and lots of instances available to choose from right now, mitigating the negative impact of any small number of instances being censored. By contrast, ATProto has one thing that looks a lot like an instance..I think given that, it's fair to say AP more decentralized than ATProto?
(DIR) Post #AxWCnBmDMEvNmQmdrk by xgranade@wandering.shop
2025-08-24T04:39:42Z
0 likes, 0 repeats
@bnewbold I'm not trying to be a dick here, really trying to understand in what sense ATProto is decentralized in practice *right now*, as viewed by reasonably typical social media users. It sounds from your description like it really isn't? There's some promising stuff out there, but if it's all "pick an instance and ask for an invite key," it's really down to how overrepresented the large instances are.That's a problem on the fediverse, obviously, but sounds strictly worse on ATProto?
(DIR) Post #AxWCnCq9P5fR4vdIdk by bnewbold@social.coop
2025-08-24T05:32:12Z
0 likes, 0 repeats
@xgranade the way I phase this is that the fediverse network is definitely more *materially* decentralized. We could come up with a number like Gini coefficient, and fediverse is doing way better day to day.but that is a different question from "is the atmosphere inherently centralized".I'm not asking fedi folks to trust Bluesky or invest deeply in atmosphere stuff. But I do wish the broader fediverse would look more closely and consider borrowing pieces and collaborating vs fighting
(DIR) Post #AxWCnE7CfAtz26cQrI by xgranade@wandering.shop
2025-08-24T05:41:23Z
0 likes, 0 repeats
@bnewbold No, and that's fair. Gini coefficients seem like a decent way to measure that, agreed — m.s being so huge is indeed a problem (no shade to anyone on there, it's a systemic issue).With respect to trust and collaboration-v-fighting, though, I have to say that trust is built? AP pretty far predates AT, but bridging/porting/interop weren't day 1 features? Plus, recent transphobic moderation decisions deeply strain trust, I think, to the degree that ATProto is seen as being Bluesky.
(DIR) Post #AxWCnFEKWACGUUxdbc by ricci@discuss.systems
2025-08-25T05:30:25Z
0 likes, 0 repeats
@xgranade @bnewbold I got very interested in this idea of using Gini coefficient. I eventually came to the conclusion that Gini itself is probably not quite right, but here's what I found.Using total user numbers for about 27k fediverse servers from https://fedilist.com/instance (I'd rather use MAU but this list doesn't track it and fedidb doesn't give access to all its data), I filtered out some of the software that is known to lie about user counts (misskey, gotosocial, NodeBB), and ended up with a Gini coefficient of 0.9876 for the fediverse - close to perfectly unequal.The closest thing I am able to find for the atmosphere is this list which just looks at PDSes: https://blue.mackuba.eu/directory/pdses . I realize that isn't an apples to apples comparison but I think it's the most generous one since from my understanding PDSes are by far the most popular bit of infrastructure (not counting custom handles) to host today. The total number of accounts across all PDSes is 37,201, or 5,742 if you exclude bridgy. With ~38M users on the main bluesky, I don't need to calculate this to know that it is indistinguishable from 1.0 :) - perfectly unequal. But: Gini index aims for a fully equal distribution. I don't think that's the right target for a a protocol supporting social networks, where I think that a distribution that models typical population statistics would be more appropriate. I'm thinking maybe the parameters of a fit to the Pareto distribution, but I need to think this through more.
(DIR) Post #AxWCnH7fTu3YMPrLV2 by bnewbold@social.coop
2025-08-24T03:41:33Z
0 likes, 0 repeats
@xgranade another answer, which I love, is that folks could also just create fediverse accounts and use the bridge!in the near future they might even be able to migrate their account through the bridge:https://blog.anew.social/bounce-a-cross-protocol-migration-tool/
(DIR) Post #AxZy2dtGOOXEdXpaWO by bnewbold@social.coop
2025-08-27T01:04:18Z
0 likes, 0 repeats
@ricci @xgranade yup! there are probably economic measures of market power concentration which could be a better fit. I think something logarithmic probably makes sense: going from 99% to 98% is a much bigger change in power dynamics than 64% to 63% (for example).I'd love to have a dashboard of various ecosystems and their concentration over time. Email providers, browser share, git forge hosting, CDN domain fronting, mobile telcos, app stores, etc.
(DIR) Post #Axa4MN53CTsG1D8RH6 by ricci@discuss.systems
2025-08-27T02:15:09Z
0 likes, 0 repeats
@bnewbold @xgranade Great idea! I found the Herfindahl–Hirschman index, which is used by regulators to calculate concentration in an industry when determining whether it has lots of competition, is monopolistic, etc.https://en.wikipedia.org/wiki/Herfindahl%E2%80%93Hirschman_indexUsing the two data sources listed above, I get:For the fediverse, I get an HHI of 0.0518 - according to the wikipedia page, this would be an "unconcentrated industry", one step more concentrated than a "a highly competitive industry".For the atmosphere (adding a 38M participant to the list above to represent bluesky itself), I get an HHI of 0.9991, which is basically a pure monopoly.Do you know if I'm using an appropriate source for the atmosphere numbers? (If you think this is likely to be a good one, I will ping the author on bsky to see if I can get it in a more easily parsable form, and/or if he has any historical data)I will try to throw up a quick page to keep the numbers for these sources up to date; I really like the idea of doing similar for the other kinds of hosting services you mention, though I don't know if it will be possible to get numbers that are as high-quality - one very nice thing about federation is how easy it is to crawl most of the network :)
(DIR) Post #Axa4lReZKre3RNhkO0 by xgranade@wandering.shop
2025-08-27T02:19:39Z
0 likes, 0 repeats
@ricci @bnewbold Huh, that metric seems to be basically one over the effective sample size, a metric for measuring how many effective samples you have in a distribution when those samples have unequal weight. So like that metric would tell us that *effectively* the fediverse is shaped the same way as a hypothetical fediverse with 20 uniformly sized instances.That's... way smaller than I would have thought.
(DIR) Post #Axa5Ym8lxAobAbM31M by xgranade@wandering.shop
2025-08-27T02:20:24Z
0 likes, 0 repeats
@ricci @bnewbold (It should go without saying, though, 20 is still a lot larger than the industry standard of 1, but does speak to that there's a long way to go in terms of forming resilience.)
(DIR) Post #Axa5YnCi01YeT6ChnM by ricci@discuss.systems
2025-08-27T02:28:35Z
0 likes, 0 repeats
@xgranade @bnewbold Yeah, I do wonder if there is a different metric here that could be used for resilience - eg. m.s going down would take down about 25% of MAUs; I was wondering about an index that counts up how many instances have to go down to take down half of the MAUs or something. I'll keep looking for measures like this (instead of making up my own :) ) FWIW, here is some work I did on this on an instance and network level a while back:https://discuss.systems/@ricci/114396317436420669
(DIR) Post #Axa5gxBokzV9krg8em by xgranade@wandering.shop
2025-08-27T02:30:04Z
0 likes, 0 repeats
@ricci It's hard to find a good proxy for "how many people does a fascist need to threaten in order to erase an oppressed group from the network," to be sure.
(DIR) Post #Axa6YaxdcQPUS2cmJs by ricci@discuss.systems
2025-08-27T02:39:47Z
0 likes, 0 repeats
@xgranade A few points of interest in that very, very long thread:An analysis of user-level decentralization found that the fediverse is significantly more decentralized in 2025 than it was in 2022: https://discuss.systems/@SystemsAppr/114396653191232099 and https://aus.social/@Drbruced/114335400872850185 - this suggests that despite the large size of m.s and the UI changes that centered it more in the signup process, we are actually heading in the right directionSome other interesting musings: https://neuromatch.social/@jonny/114396706353836320
(DIR) Post #Axa6lTHlT0fQPYB7Ts by ricci@discuss.systems
2025-08-27T02:42:06Z
0 likes, 0 repeats
@xgranade I also wrote up some thoughts on the vulnerability of the DNS in particular (though without numbers): https://discuss.systems/@ricci/114412296355937121
(DIR) Post #B0OLuc8qmDMo4L2fjs by ricci@discuss.systems
2025-11-19T04:11:35Z
0 likes, 0 repeats
@bnewbold @xgranade Reviving an old thread because I think I have got pretty close to building the dashboard you theorized about, @bnewbold.net - I've got good data from the fedi, atmosphere, web hosting, git forges, email services, DNS providers, and CA. I was somewhat gratified to see an academic paper come out in September that (independent of me) makes the case for using the HHI in this context. I am also using the Shannon Index because, as you pointed out, logarithmic scale helps you see the changes in the smaller participants more clearly.https://arewedecentralizedyet.online/In case you wonder why the Atmosphere shows as "centralizing" right now: that is not (AFAICT) an indication of users moving *back* to Bluesky PDSes - it's just that these PDSes grow so much faster than any of the alternatives that what happens is you see it decentralize when there are bursts of people moving to blacksky, etc., then it swings back when those level off and the bluesky ones keep growing.