bryan newbold

oscilloscopes, cycling, snow, big cities, wiki. I like speculating about found objects.
protocol and standards @bsky.app. formerly archive.org
elsewhere: bnewbold.net / @bnewbold@social.coop

🌉 bridged from 🦋 bnewbold.net, follow @bsky.brid.gy to interact

Public Key

npub1tull74k2f75pwlc39yg2xn6tve9k6g854yjxafuu4x9rqggcnwas6tf77h

NIP-05 Address

bnewbold.net@bsky-brid-gy.mostr.pub

Profile Code

nprofile1qqs970ll2m9yl2qh0ugjjy9rfa9kvjmdyr62jfrw57w2nz3syyvfhwcpz4mhxue69uhhyetvv9ujuerfw36x7tnsw43qtmx60n

Publishing to

relay.ditto.pub

Author Public Key

npub1tull74k2f75pwlc39yg2xn6tve9k6g854yjxafuu4x9rqggcnwas6tf77h

Show more details

Last Notes

2026-04-07 05:37:32 UTC

devs working with bsky image embeds:
we are planning to bump the image embed blob max size from 1 MByte to 2 MByte. first in the lexicon schema, and then in the Bluesky Social app.
more about impacts, and what this means in terms of lexicon evolution, in this github discussion
Increasing Bluesky Post Image ...

2025-11-02 00:45:24 UTC

"composable moderation" is about the ability to use multiple mod services at the same time, and that mod services can be operated independent of other infra (aka service modularity)

2025-11-02 00:39:49 UTC

but I also want students/hackers to use the tools they have at hand to *read* data, and in a lot of situations that will be a generic CBOR library which doesn't enforce any special encoding rules

2025-11-02 00:38:07 UTC

I know we don't want one bad *record* to break an entire *repo*.
I think PDS should be very strict, ours maybe is not, and that is urgent to fix.
I would love if every SDK and service consistently did full validation and consistently rejected bad records...

2025-11-02 00:35:35 UTC

but in our IETF text we should only reference formal standards, so we will ref CBOR RFCs and re-describe details.
still no floats.
more ambiguous question is where all this is expected to be validated, and behaviors if it fails. PDS, relay, SDKs, etc

2025-11-02 00:12:20 UTC

but yeah have wanted an "Automated Account" indication forever

2025-11-01 20:43:57 UTC

aka there are definitely groups of 8 friends I would care more about being excluded from that 800k strangers

2025-11-01 19:36:43 UTC

if your big receipt is somebody maybe going to a big meetup with Bad People long ago, then it sounds like you don't have much of a critique at all.
which is disappointing because we do make mistakes and need effective critique to hold us accountable.

2025-11-01 18:25:14 UTC

probably also some affordances around "hey you have been using this list for 6 months and it grew 10x and now 8 people you follow are on it. you sure?"

2025-11-01 18:15:38 UTC

it is 2025 and I have been on team "augment not automate" for almost 20 years

2025-10-31 06:46:01 UTC

*don't

2025-10-31 06:03:10 UTC

check out @stechlab-labels.bsky.social

2025-10-30 22:37:19 UTC

pretty interesting protocol-level lexicon data validation scenario! we knew this would happen eventually, and now it has
Grapheme count mismatch due to...

2025-10-30 21:43:42 UTC

damn week two and he's already making up words
Week 2: Or, It's actually a bl...

2025-10-29 23:47:19 UTC

we absolutely intend to treat Blacksky-hosted accounts equally to Bluesky-hosted on our infrastructure, including the relay and appview. I'm sorry that didn't happen here.
the meta-failure is that we don't have automated alerting for this scenario.

2025-10-29 23:44:51 UTC

it looks like we may have accidentally failed to bump rate-limits on the primary bluesky relay, causing a bunch of (legitimate) accounts to not have events get processed.
those limits have been bumped way up, and I will follow up soon to ensure this sort of thing doesn't happen again.

2025-10-27 17:52:20 UTC

I wish I had elaborate costumes for certain types of programming.
like, oh, newbold is wearing that traditional samurai outfit at his desk again, I guess he's resolving a big merge conflict

2025-10-27 02:07:51 UTC

last night was "Devil May Caraway" and it was banger

2025-10-26 05:08:19 UTC

@retr0.id wait does this work well?
if so changes my vibe on USB C a bit. can you do power? what kind of transfer speed?
GitHub - AnasMalas/pcb-edge-us...

2025-10-25 19:33:20 UTC

(doesn't get "rejected at relay" or anything, just treated as lexicon-non-compliant, so appviews would ignore)

2025-10-25 16:26:27 UTC

the reason for the future thing is that the sensitive are that once a message is validated, no message with an earlier TID will be accepted. we didn't want an accidental message from 200 years in the future to pseudo-break the account

2025-10-25 16:24:31 UTC

a future TID is still a totally valid TID for what it is worth, was just sitting that some library might be strict in that peculiar way

2025-10-24 20:12:28 UTC

but I suspect it will be a fork, new project, or maybe open sourcing of an existing project (eg, something might emerge from ROOST). TBC I think Bluesky can/should/will develop or fund this work.
but it is in contrast to, for example, the PDS reference implementation

2025-10-24 20:09:44 UTC

the main reason for that is that Ozone has long since passed over a threshold in to "enterprise software" territory, out of "community open source".
the community does desperately need a "community open source" solution in this space...

2025-10-24 20:07:53 UTC

that would make a lot of sense, but would require Ozone to prioritize multi-stakeholder governance: it would need to be responsive to the needs of other users and applications and clients.
and unfortunately for that specific project I don't see the path forward to that. it would need a fork.

2025-10-24 20:02:54 UTC

this is confusing: the "ref" mixes semantics with JSON structure layout. we should probably have "app.bsky.labeler.service" simply re-define the JSON structure instead of referencing "com.atproto.label.defs#selfLabels"

2025-10-24 20:02:12 UTC

what it has is a field ("labels") which *happens to the same JSON structure* as some other separate data structures which have to do with "self labels". instead of re-defining that JSON structure, the field uses a "ref" over in to that existing schema definition.

2025-10-24 19:22:10 UTC

in contrast to labeling, which is a protocol-level concept.
eg, some systems (IIRC Mastodon) allow creating a single report which references multiple pieces of content. the current mod reporting system doesn't do that. we shouldn't say "atproto doesn't support that", just "ozone doesn't"

2025-10-24 03:50:06 UTC

(am I doing this right)

2025-10-24 03:49:43 UTC

is there a Discourse yet that AI glasses are anti-accessibility because they will make people distrust folks with regular spectacles

2025-10-23 23:54:56 UTC

maybe generic infra like relays, mirrors, generic appviews (eg, microcosm).
product design is hard, and i'm not sure regular grant cycles are a good fit.
it would be cool if governments gave out "open social vouchers" which citizens could give to local projects they support. or matching funds?

2025-10-23 23:52:22 UTC

I think the most fund-able bits are baseline moderation (abuse, spam, botnets; the less controversial stuff) and SDK development (eg proto implementations, security audits, performance, compliance tests).
maybe resources for auto-classification, translation, accessibility?

2025-10-22 05:34:20 UTC

wonder if erowid traffic and community participation are down due to AI sumarization

2025-10-22 04:49:21 UTC

put out a new 'goat' release (v0.1.2) with basic support for PDS administration ('goat pds admin ...').
if you run it from a PDS instance it should find admin auth info automatically.
hope to have this replace the 'pdsadmin' shell script over time

2025-10-20 08:50:30 UTC

kind of embarrassed how helpful this is
#nevent1q…qtru

2025-10-20 08:36:36 UTC

there is an AWS incident happening in us-east-1:
health.aws.amazon.com/health/status
Bluesky has been mostly unaffected so far, but automated emails may be impacted, including communication of moderation actions

2025-10-20 02:15:47 UTC

this has actually happened, eg bluesky re-ingested our own labels from scratch a could times IIRC

2025-10-18 17:58:27 UTC

sorry for tone, it just keeps coming up

2025-10-18 17:57:55 UTC

@clearsky.app could you fix your shit?
these are not real accounts, and doing a ranked list like this keeps freaking people out. if you want accurate counts you should check against PDS, or just query the PDS for a list.
most of the morel counts are also fake, and the plc.surge.sh ones also fake.
#nevent1q…grd7

2025-10-18 17:32:05 UTC

(uncoordinated meaning not prepared, but driven by the posts on twitter)

2025-10-18 01:00:11 UTC

CTO: So are we clean on OPSEC?
Backend: Sure are boss, real fuckin sexy for #FCF, just like you asked.
CTO: what

2025-10-17 23:26:26 UTC

what is everybody up to this weekend? any big events going on?

2025-10-17 22:13:29 UTC

as a reminder: bluesky (the app) has a broad set of user interactions controls.
beyond blocking and block lists, we have features to control who can reply, mention, and quote-post accounts
#nevent1q…hzgt

2025-10-17 19:46:08 UTC

ban evasion (creating new accounts on any instance) is a perennial challenge on all open-signup systems, including centralized platforms. same with brigading (coordinated groups of ban evasion).

2025-10-17 19:44:42 UTC

aka, labels and actions at the appview are sticky to the account DID, not their hosting location.
if you are using the bsky app and getting notifications from a user who the bsky mod team has taken down, that is a bug we need to fix (you should not get notifs in that situation)

2025-10-14 19:33:51 UTC

a specific important case of this is appeals.
on the other hand, I definitely think that labelers need to be able to back-pressure reports: they should not need to accept an unlimited number of reports of any kind from any account

2025-10-14 00:45:20 UTC

better link
youtu.be/X2fGzmphx4U?...
Boston bomb scare suspects pre...

2025-10-11 20:48:50 UTC

the bsky appview self-hosting story isn't great, but folks have gotten it working in the past, and there are differing schools of thought on what would be best for independent hosting: big collective servers? small pared-down UX? generic app frameworks?

2025-10-11 20:45:34 UTC

Valenciaga St

2025-10-11 00:16:17 UTC

much lower-stakes to fiddle or cut corners with vs mod affordances (labels, blocking)

2025-10-10 22:46:58 UTC

a complexity is handling account security / takeover scenarios, in which case you didn't want to allow migration.
also the "adversarial" scenario still needs to be more accessible, and maybe that should be prioritized first

2025-10-10 21:32:00 UTC

standards folks!
if you are in to this sort of thing, we now have an IETF mailing list to discuss AT protocol bits.
mailman3.ietf.org/mailman3/lis...
if you want to jump in, I recommend setting up an overall IETF datatracker account and reviewing IETF IP policies ("note well") first

2025-09-10 03:47:17 UTC

to clarify if there is any confusion, Bluesky Social PBC does not currently own or control 'bluesky.app' (the domain with full "blue" in it)

2025-08-14 16:52:22 UTC

AI scraping volume is centralizing the web by making "hosting a website" as difficult as "hosting your own email".
even if you are fine having your site crawled, the scrapers are so aggressive that you basically need to use a CDN or tool like Anubis or be able to serve millions of req/day

2025-06-13 22:32:08 UTC

this is one of the reasons we strongly push folks to set up a *.bsky.social handle to start (at least on our servers), then update it to a custom domain handle. the onboarding flow kind of sucks otherwise

2025-03-08 01:34:44 UTC

👩‍💻 ¡atproto devs! 🧑‍💻
just in time for the weekend: our protocol roadmap for the first half of 2025.
we got auth scopes, we got merkle tree inversions, we got deprecations, we got balmy spring weather and long walks in the forest. check it out!
2025 Protocol Roadmap (Spring ...

2025-02-20 19:59:19 UTC

🛠️ atproto devs! 🛠️
got a new proposal out about Sync 1.1, aka inductive firehose, aka cheap validating relays.
it is a long, low-level doc. doesn't directly impact folks unless they emit a firehose (eg, PDS and other repo hosts), or are doing full validation as a consumer
proposals/0006-sync-iteration ...