geoffrey richard_

Quietly this year, the protocol under Bluesky started leaving Bluesky. The IETF chartered a working group, "Authenticated Transfer Protocol", to standardise atproto's core. Worth understanding what that does and, more importantly, what it doesn't. 🧵
---
What it changes is governance. The repo format, the sync protocol, the at:// URI scheme: these move from Bluesky's reference implementation to vendor-neutral IETF specs. Bluesky's own framing is that the core protocol gets "a home outside of and independent from Bluesky PBC."

First caveat: this is early. The deliverables are Internet-Drafts, not finished RFCs. An I-D is a working document: it can change completely or expire. The charter was approved in early 2026; the actual standards are a multi-year job. "Standardised" is the destination, not the current state.

Second caveat: it's narrow. In: repository structure, sync, the at:// scheme, account-identifier resolution. Explicitly out : private data, the app.bsky lexicons, and the whole labeling/moderation system. The parts you actually feel as a user remain Bluesky-driven. For now.

Also the live network still leans on one company. Main relay, primary AppView, did:plc, the default PDS — all Bluesky PBC. You can self-host every piece, but the defaults are theirs. The architecture is decentralised; the network isn't yet. Does standardizing the specs actually change that?

I have high hopes for the at-proto. I’m actually hosting my own PDS so my data is hosted by me and authenticated by me. I also anchor my identity with my domain name, geofox.org, with a did:web, instead of relying on Bluesky’s Public Ledger of Credentials (that governance is actively being moved out to an independent association).

That domain is linked to my identity for years, so it was the logical choice. I’ve published a bit more info on my blog : geofox.org/posts/decentralized-social/#bluesky--at-protocol-my-data-my-domain-my-did

On the other side of the social spectrum, Meta's ActivityPub support on Threads isn't openness, it's an alibi. Federation is opt-in, buried, and one-way for everything that matters: your replies and graph don't come with you, so there's no door out, only a window in. They adopted the wire format and skipped the portability, the one part that would actually let users leave. That's not interoperability but a compliance gesture dressed as principle, shipped by the company with the most to lose if it were real.

Sources: IETF charter (Authenticated Transfer Protocol WG) datatracker.ietf.org/doc/charter-ietf-atp/ · Bluesky's announcement atproto.com/blog/kicking-off-the-atp-working-group

🌡️ Europe is under a heatwave, let’s talk about sunscreen! Many think mineral sunscreens are "natural" and chemical ones are toxic, but (surprise!) it’s not the case at all. A thread 👇

🍀 The "Natural" Myth: Mineral sunscreens (Zinc/Titanium) aren't just crushed rocks from the earth. They require intense synthetic processing and synthetic coatings (like silica) to stop them from generating dangerous free radicals when exposed to sunlight.

🪞 The Reflection Myth: People claim mineral filters "reflect" UV like a shield. As Lab Muffin Beauty Science points out, mineral filters actually protect skin by absorbing almost all of UV rays, working almost exactly (barring a few percent of reflection) like chemical filters.

👨‍🔬 Toxicology: What about chemical endocrine disruption? Lab Muffin explains that animal studies showing harm use massive, physically impossible doses. FDA blood absorption limits are default regulatory triggers for more testing, not toxic danger thresholds.

🇪🇺 US vs EU: The US FDA regulates sunscreens as OTC drugs, meaning no new filters have been approved since 1999. The EU treats them as cosmetics, allowing over 30 advanced chemical filters (like Tinosorb) that offer vastly superior, photostable UVA protection.

💡 Bottom line: Because mineral sunscreens leave a thick white cast, people apply way too little, drastically lowering their UV protection. Modern EU chemical sunscreens are cosmetically elegant and highly effective, making them the superior choice for real-world use.

📚 See : Wong M. How much UV does a REAL mineral sunscreen absorb and scatter?. Lab Muffin Beauty Science. June 28, 2021. Accessed May 16, 2026. labmuffin.com/how-much-uv-does-a-real-mineral-sunscreen-absorb-and-scatter/

Let’s talk about fake science. One of Professor Dave’s latest is a takedown of the alt-health/supplements business that Joe Rogan and the MAHA crowd peddle. Well done and funny, as usual. youtu.be/wwLdFrjsN_c

In Western Europe a bit of regulation gives us cover, but homeopathic sugar pills still sell here. Quick primer on why homeopathy isn’t medicine. It was invented in 1796. before atoms, before germ theory, before modern chemistry.

The premise: a substance that causes a symptom in a healthy person cures that symptom in a sick person, if you dilute it enough. Now the dilution. A standard “30C” preparation is 30 successive 1:100 dilutions — 1 part in 10^60.

Past 12 dilutions, statistically zero molecules of the original substance remain. You’re drinking water. The pills are lactose. Defenders invoke “water memory.” There’s no known mechanism : water’s hydrogen bonds reorganize in picoseconds. Australia’s NHMRC reviewed 225 trials in 2015: no condition with reliable evidence of effect. Europe’s EASAC reached the same verdict in 2017.

This matters because sugar pills aren’t neutral when people skip real treatment. In 2017 a 7-year-old in Italy died of an ear infection treated with homeopathy instead of antibiotics. France ended reimbursement in 2021, the UK NHS in 2017. It’s still on pharmacy shelves across Western Europe. Here, in Belgium, it’s still reimbursed by some insurance fund. 230 years of confidently selling nothing.

Is Musk planning a “rapid scheduled disassembly” of your savings?

SpaceX reportedly wants a ~$1.75T valuation on $18.7B revenue and a $4.9B 2025 net loss: roughly 94x sales.

That’s a hardware + AI capex story being priced like a near-monopoly software platform.

Even peak-2020 “expensive” tech IPOs traded at a fraction of that multiple. xAI reportedly burned ~$14B in 2025 while generating only ~$3.2B revenue, and AI infrastructure now dominates company capex. Public investors are effectively being asked to fund Musk’s AI ambitions at SpaceX valuation levels.

And that’s before considering the super-voting share structure that leaves Elon Musk with overwhelming control and public shareholders with very limited influence.

Maybe it works. There is a dominance of Starlink and launch monopoly characteristic. But personally, I’m not taking that bet.

“The open web is on its way out,” “Internet traffic now often begins and ends with a visit to Google.” “With A.I., Google is reducing everyone to raw data providers.”

Not to be a Luddite but it’s sad.

nytimes.com/2026/05/19/business/google-seach-bar-ai-gemini.html