Let’s Encrypt has been down most of today

Let’s Encrypt’s production API experienced degraded performance for most of June 18-19 due to networking issues between its two high assurance data centers, with clients encountering 400 and 500 error responses. The team stabilized performance with reduced redundancy while investigating upstream ISP problems. Commenters noted the irony of an entity pushing for shorter certificate lifespans suffering a prolonged outage, and one user reported an IoT vendor already serving expired certificates as a result. Several pointed out that Firefox treating recently expired certificates the same as MITM attacks makes these outages more painful than necessary.

Ice Water Drowning Survival After 147-Minute Submersion

A case study published in JACC details a patient who survived 147 minutes of submersion in ice water followed by hypothermic cardiac arrest, a scenario where cold-water immersion paradoxically preserved brain function long enough for successful resuscitation. Commenters discussed how medical training has evolved on shockable vs. non-shockable rhythms, with one noting you can’t shock an asystole heart. Others wondered whether the protocol could improve survivability for regular (non-hypothermic) drowning cases.

To study how chips work, MIT researchers built their own operating system

MIT researchers built a custom operating system to run experiments on modern processors at a level of detail sufficient for studying attacks like Spectre and Meltdown, rather than relying on opaque commercial OS layers. The work targets the gap between hardware behavior and what software can observe. A commenter familiar with XNU’s build process noted that the difficulty isn’t redaction but toolchain compatibility, arguing the paper’s citation of darwin-xnu-build doesn’t support its claims. Others speculated about applications for controlled environments like game consoles.

How Japan’s railways stayed one while splitting apart

After Japan National Railways was privatized and split into seven JR companies in 1987, the system maintained a unified identity through a shared logo, coordinated scheduling, and interoperable ticketing — despite being independently operated. The article traces how the visual and operational cohesion was preserved across the fragmentation. A Japanese commenter stressed that JR represents only 7 of roughly 100 train companies in Japan, cautioning against drawing conclusions about the entire network from JR alone. Others noted the JR logo resembles the kanji 駅 (station), possibly an intentional design choice.

Zero-Touch OAuth for MCP

The Model Context Protocol published its Enterprise-Managed Authorization extension, enabling organizations to centrally manage authorization for MCP servers so that agents inherit credentials without per-user OAuth flows. The spec is now stable, with early adopters including Okta, Microsoft, Figma, and Linear. Commenters emphasized that MCP’s real advantage over skills/CLI is isolating auth flows outside the agent’s context window. The new token format, ID-JAG, is being tracked through IETF for standardization.

DuckDB Internals: Why Is DuckDB Fast? (Part 1)

Greybeam published a deep dive into DuckDB’s architecture, tracing its evolution from a CWI Amsterdam research project in 2019 to one of the most widely adopted databases of the past decade. The article covers the vectorized execution engine, columnar storage format, and design decisions that make DuckDB performant for analytical queries. Commenters praised DuckDB’s ease of use as the primary adoption driver — running select * from 'data.json' works out of the box — with its speed and versatility keeping users locked in.

Building a robotics research setup that lives next to my desk

DFDX Labs documented a desktop robotics research setup with multiple camera feeds, a human operator interface, and real robot hardware that’s now affordable enough for individuals and small teams. The post covers the hardware choices, control pipeline, and why the author intentionally skips camera calibration. One commenter who built a similar system with a Franka arm noted the challenge of doing multi-view calibration solo. Another observed that teleoperation robots remain surprisingly slow despite years of development.

Datasette Apps: Host custom HTML applications inside Datasette

Simon Willison launched datasette-apps, a new Datasette plugin that lets you host custom HTML applications directly inside a Datasette instance, keeping apps and data co-located. The plugin leverages modern browser APIs for safe HTML rendering without heavyweight dependencies. Commenters appreciated the pattern of keeping application code next to data, with one noting they’d previously hacked together similar functionality using Datasette’s JSON endpoints and separate HTML pages. The built-in browser HTML sanitizing was highlighted as a significant improvement.

Cell-based architecture for resilient payment systems

American Express engineering detailed their cell-based architecture for payment processing, where the system is partitioned into isolated cells that can fail independently without taking down the entire payment pipeline. The design targets high availability and predictable performance for live transaction processing. Commenters drew comparisons to Erlang supervision trees and nanoservices. Some felt the post was vague about what the middle layer actually does — no detail on routing, data sync, or transaction isolation — making it read more like a scaling story than a payment technology deep dive.

Zork name origin got an update on Wikipedia

After two years of research, a blogger traced the etymology of “Zork” — the classic text adventure’s name — and got the Wikipedia entry updated with new findings about its likely derivation from MIT hacker slang. The article walks through the investigation process, including comparisons to earlier research by Nick Montfort. Commenters found the edit distance between “work” and “zork” (just one letter) compelling, noting the MIT convention of replacing first letters with “z” for placeholder names. The tab-completion theory also drew amusement.

Horizons JPL Solar System Data Demo

A developer built an interactive demo combining NASA JPL’s Horizons solar system data with live NASA Deep Space Network telemetry, using Datastar and Common Lisp. The interface lets users click planets to see JPL Horizons details while DSN antenna logs update in the background across all stations. A commenter praised the 2D sprite TUI approach to visualizing a JPL telnet service as exactly the kind of niche tool the internet was built for.

I told them forced consent was unlawful. 5 years later it cost Elkjop €1.8M

A privacy advocate’s five-year battle against Norwegian electronics retailer Elkjop (Elkjøp) culminated in a €1.8 million GDPR fine for forcing customers to join a marketing club as a condition of receiving offers. The blogger had documented the violation in writing early on, which proved critical to the eventual enforcement. Commenters applauded the persistence and noted the importance of getting violations documented in the company’s own words. Discussion touched on how rare this outcome is, especially compared to US consumer protection.

Show HN: Are You in the Weights?

“In The Weights” is a tool that lets you check whether your name appears in major AI training datasets, showing which attributes and contexts are associated with your identity in model weights. Users can search their name and see how LLMs have encoded information about them. Commenters reported mixed results — some found themselves accurately described while others collided with celebrities or historical figures. One user’s name matched six soccer players yet still ranked first. Many appreciated the tool while refusing to use their real names.

The Token Compression Illusion: Why I’m Skeptical of RTK

A software engineer published a critical analysis of RTK (a tool that compresses terminal output for LLM agents), arguing that compressing context for agents introduces structural flaws that undermine reliability and operational safety. The article frames RTK as part of a broader pattern of “LLM magic box” tools that sound too good to be true and usually are. The author clarified the motivation was management pushing cost optimization without understanding accuracy tradeoffs. Commenters agreed benchmarks for this space are extremely difficult due to the combinatorial explosion of harness, model, and CLI combinations.

If your product is Great, it doesn’t need to be Good (2010)

Paul Buchheit’s 2010 essay argues that products like the iPad succeed by being genuinely great at one thing rather than competent at many — being “good enough” across the board is a recipe for mediocrity. The post uses the iPad’s missing features as evidence that excellence in a core experience trumps feature completeness. Commenters compared this to the original Walkman, where engineers wanted recording functionality but were told no to preserve the singular purpose. Others noted the hardest part of product management is saying no to reasonably good ideas.

Noam Shazeer Joins OpenAI

Noam Shazeer, co-author of the seminal “Attention Is All You Need” paper that introduced the transformer architecture, announced he’s leaving Google to join OpenAI. Shazeer had returned to Google relatively recently through the Character.AI deal, making this another surprising departure. Commenters linked to Wired’s backstory on the eight Google employees who invented modern AI, providing context on each author’s contributions. One commenter speculated about what could have caused him to leave so soon after returning.

Show HN: Talos – Open-source WASM interpreter for Lean

Cajal Technologies released Talos, a WebAssembly interpreter written in Lean, designed specifically for formal reasoning about program execution. The project bets that Wasm is a strong verification target since many languages compile to it, and Lean is the right language for writing proofs about that execution. A commenter noted the name collision with Sidero Labs’ Talos OS, causing brief confusion about native Wasm pod support in Kubernetes. The Cajal team clarified they’re interested in feedback on whether specs should be written in Lean against Wasm semantics or by annotating Rust directly.

Agentic Resource Discovery Specification

A new specification proposes a standard way for AI agents to discover and use external capabilities — tools, Skills, MCP servers, APIs, workflows, and other agents — addressing the growing problem of agent resource proliferation. The spec defines a searchable registry with multi-protocol support across MCP, A2A, and other emerging standards. Commenters saw overlap with MCP’s existing Resources concept, with the main addition being multi-protocol search. One practitioner described solving the same problem with a Wiki plus saved programs approach, finding ontology construction fundamentally unsolved even within a single company.

The founder of Craigslist has given away half a billion dollars

Craig Newmark has donated over $500 million through his philanthropic foundation, focusing on journalism, cybersecurity, and veterans’ causes. The article highlights his fear that American generosity is increasingly under attack from political trolling. Commenters recalled interviews where Newmark described living modestly despite his wealth. Some pointed out that Craigslist’s minimalist design and resistance to “improvement” was itself a form of generosity, though others estimated 25% of apartment listings are scams and argued better moderation could have done more social good than the donations.

Migrating from GNU Stow to Chezmoi

A developer documented switching from GNU Stow to Chezmoi for dotfiles management after symlink management across multiple devices became unwieldy. The post covers Chezmoi’s templating system, file naming conventions, and how it handles machine-specific configurations without symlink headaches. A Stow defender argued that write-through behavior is a feature, not a bug — git immediately shows dirty files when something changes, giving you the choice to commit or revert. Others recommended alternatives like Nix Home Manager or mise’s new bootstrapping features.

Midjourney Medical

Midjourney announced a medical imaging initiative exploring AI-enhanced tomography using phased array beamforming techniques, aiming to reduce radiation exposure and cost compared to conventional CT scans. The announcement takes an unusually philosophical tone, asking fundamental questions about what healthcare technology should become. A practicing radiologist found the modality promising but cautioned about over-detection creating unnecessary follow-ups. Commenters with imaging experience questioned how image generation expertise transfers to a fundamentally different imaging problem, and whether nanometer deflection sensitivity translates to clinical image resolution.

Update on Ocean Observatories Initiative

The NSF announced it will not proceed with dismantling the Ocean Observatories Initiative after the Senate passed the Saving the OOI Act with unanimous consent, prohibiting the program’s elimination. The reversal follows widespread scientific community opposition to proposed budget cuts. Commenters noted this represents a retreat from impoundment — the administration’s stance that not paying congressionally authorized funds is constitutional — and required real political pressure. The bill still needs House passage.

Ask HN: Is anyone using the A2A protocol?

An HN user asked whether Google’s Agent-to-Agent (A2A) protocol has seen real-world adoption six months after launch. The discussion reveals minimal practical usage, with one developer noting A2A targets enterprises — Google recently rebranded their AI suite as “Gemini Enterprise Agent Platform.” Another developer evaluated A2A carefully but concluded it doesn’t solve the correct problem because it makes no distinction between agent types. Laurie Voss’s skeptical presentation on agent interaction protocols was cited as evidence that A2A, MCP, and similar protocols may not add enough value to justify the complexity.

The Korean telecom giant at the center of Anthropic’s Mythos controversy

Wired reports on SK Telecom’s role in Anthropic’s Mythos controversy, tracing the Korean telecom’s $100 million investment in 2023 and subsequent commercial partnership, which intersected with White House export control concerns. The article examines whether the SK Telecom relationship was a genuine security concern or a convenient pretext. Commenters argued the Wired headline reframes the issue misleadingly — SK Telecom was a previously resolved issue, while the actual trigger was industry reporting around the Fable launch. Others saw the broader pattern of administration actions against AI companies as politically motivated.

I need your clothes, your boots, and your motorcycle

MAME developer R. Belmont posted a progress update on Power Macintosh emulation, detailing the challenges of debugging code written in three languages (PowerPC, emulated 680×0, and compiled FORTH). The post covers the new PowerPC DRC (Dynamic Recompilation Core) that translates PowerPC machine code into native host code for better performance. Commenters linked to an agentic retro-gaming series where AI was used to rapidly develop a cycle-accurate Intellivision emulator. One user complained the Terminator-referencing title gave no hint about the actual content.

Dutch Railways offers unlimited off-peak train travel nationwide for €49/month

NS (Nederlandse Spoorwegen) launched a summer promotion offering unlimited off-peak train travel across the Netherlands for €49/month, with children traveling free. The pass covers the entire Dutch rail network during off-peak hours. Commenters noted the introductory price jumps to €127.95 after the summer period. Comparisons were made to Germany’s Deutschlandticket (€58/month, valid all day on local and regional trains in a much larger country). A Swiss commenter noted their constitution prevents making public transport free, even in cities that want to.

Show HN: Crawlie – Free open-source SEO audit tool for humans and agents

Crawlie is a fast, free, open-source technical SEO and GEO crawler hosted on GitHub, designed to work for both human users and AI agents. The tool audits websites for SEO issues and is built to be integrated into agentic workflows. This story had no HN comments at time of collection.

CBC to stop airing NHL games after 74 years

The CBC will stop airing NHL hockey after 74 years, ending the era of free over-the-air hockey broadcasts on Canadian public television. The decision marks a cultural shift as hockey moves entirely to paid streaming and cable platforms. Commenters noted that hockey participation and viewership in Canada is already declining, and losing free access will likely accelerate the trend. One linked to a documentary examining the broader decline of hockey in Canadian culture.

Notes from a tired Egyptian whose job is explaining that humans built the pyramids

McSweeney’s published a satirical piece from the perspective of an Egyptian exhausted by explaining that humans — not aliens — built the pyramids, channeling the frustration of archaeologists dealing with persistent pseudoscience. Commenters connected it to the “I personally can’t conceive of how this was built, therefore aliens” logic that underpins Ancient Aliens theories. One line resonated: “People dramatically underestimate what thousands of organized humans can accomplish when they are adequately fed, aggressively supervised, and denied alternative career paths.” The classic McSweeney’s “E-mail Addresses That Would Be Really Annoying to Give Out Over the Phone” was shared as a companion piece.

Setting Up a New Windows Laptop in 2026

A developer documented the full process of setting up a new Windows laptop in 2026, covering the 24+ steps needed to strip bloatware, configure privacy settings, and install essential development tools. The walkthrough captures the friction of starting fresh on a modern Windows machine. Commenters strongly recommended installing Firefox with uBlock Origin and making it the default browser rather than Chrome. One user noted the process is still at least 24 hours faster than setting up a fresh Ubuntu ThinkPad, which required a 70+ step document. Another shared the experience of setting up machines for a family mortgage business — bog-standard Office 365 setup that still required hours of configuration.