Today’s Highlights#

Start with the Dutch DAWO community — a government-led effort to rebuild the civil servant workplace from replaceable, auditable open-source blocks, with DAWO-NixOS as the operating system base. The site frames the goal as digital sovereignty and verifiability, with every layer inspectable and swappable and code published on code.overheid.nl and mirrored on Codeberg. Around that bid to make public IT more controllable, today splits into three parallel questions: when models like Claude reach defense and supply chains, where contract boundaries and risk labels should sit; when billionaires can move a tax base but land cannot, how California should choose between a wealth tax and a land value tax; and, closer to daily practice, how smaller technical bets — Go’s chip-level acceleration, Ollaya’s local decision models, Rails at a crossroads, and a touchable Factorio — show different paths for putting capability into tools and communities.

Policy and Governance#

The Netherlands builds a replaceable workplace on NixOS#

The DAWO.community overview describes a digitally autonomous workplace made not of one suite but of building blocks for operating system, cloud, collaboration, and AI, each inspectable and replaceable. The operating system layer is DAWO-NixOS — NixOS is a Linux distribution known for reproducible builds — presented as a reproducible and verifiable base, with blueprints and components published on code.overheid.nl and mirrored on Codeberg and with calendars and forums for joint development. The project’s reading is that buying public IT should buy long-term control: if every layer can be audited and swapped, migration costs fall and bargaining power rises. In the HN thread, many welcomed the modular, open approach as the only way to avoid trading one lock-in for another, while others warned the hard part is procurement, training, and sustained maintenance budgets. Several top comments dug into hosting and governance details, treating autonomy as an ongoing organizational investment rather than a one-time switch. Discussion: Hacker News thread.

Appeals court keeps Anthropic’s supply-chain risk designation#

CNBC reports that a 2-1 panel of the U.S. Court of Appeals for the District of Columbia Circuit upheld the Pentagon’s designation of Anthropic as a supply chain risk. The dispute traces to collapsed talks over Claude on GenAI.mil, the Pentagon’s AI platform: the department sought unrestricted use across all lawful purposes, while Anthropic sought to exclude fully autonomous weapons chains and domestic mass surveillance. Since the March designation, the military and affected contractors have been barred from using Claude on covered programs. The same set of designations had split the courts, with a San Francisco federal judge ruling a parallel track unlawful in August, and the appellate panel stayed its mandate to allow Anthropic to seek rehearing or appeal. The thread was sharply divided. Some argued that labeling a supplier that insists on use restrictions as a risk and finding another vendor is reasonable management, especially around lethal chains; others saw the label as retaliation that could ripple through the broader contractor ecosystem and called for clearer, more transparent contracting rules. Discussion: Hacker News thread.

California’s wealth tax versus the value in its land#

A widely shared essay from the Center for Land Economics dissects California’s November ballot measure for a one-time 5% tax on billionaires, to be collected over five years — roughly $20 billion a year for health and schools. The authors estimate the state’s land is worth about $8.14 trillion, roughly eight times the taxable billionaire base, and immobile. They argue the wealth tax’s assumed $2 trillion base is overstated by nearly half because six billionaires had moved residency before the January 1, 2026 cutoff and another high-profile departure could face litigation over retroactivity; to hit the same revenue with a shrunken base, rates might have to rise to 1.6% to 1.9%.Their alternative is a land value tax — a property tax that falls only on land, not buildings — at about 0.25% to raise the same sum, and they trace the fiscal hole to Proposition 13’s long suppression of property assessments since 1978. Commenters split along familiar fiscal lines. Supporters, citing Henry George, said taxing land is harder to pass on and matches the windfalls public investment creates; skeptics warned about impacts on elderly homeowners and assessment complexity, while many debated whether net worth equals collectable cash, loan-against-stock loopholes, and lessons from Norway and elsewhere. Discussion: Hacker News thread.

Tech and Products#

Go experiments with platform-independent SIMD#

The official Go blog outlines experimental SIMD (Single Instruction, Multiple Data — doing the same arithmetic on a whole vector at once for chip-level speedups) support in Go 1.26 and 1.27. Before this, using SIMD from Go meant writing assembly. The new approach first shipped a low-level architecture-specific package, archsimd, for amd64 and then for arm64 NEON and wasm, plus a portable, length-agnostic package inspired by Highway for C++. By abstracting vector length out of the type system and providing only the cross-platform intersection of operations, with efficient emulation filling gaps, the goal is write once and get near-assembly speed where available and still run elsewhere. It remains experimental, behind GOEXPERIMENT=simd. Top comments praised the two-layer design: performance seekers can drop down, most developers can stay on the portable API. Others noted everyday business code may not need hand-vectorized kernels, while databases, imaging, and crypto would benefit meaningfully. Discussion: Hacker News thread.

Ollaya brings Jev-style decision models to your own hardware#

Ollaya describes itself as Ollama for open-source decision models: ask typed questions about text or JSON and get calibrated answers in a single forward pass, with no token-by-token generation and data staying on your hardware. The page’s example runs a decider model against a request to force-push to main and returns choice and rating answers — block or not, on task or not, destructive or not — with probabilities, in about 178 ms for a 2B model on an RTX 4090. The project stresses speed, drop-in compatibility with TypeSafe’s API, open weights, and privacy by default, with desktop and command-line installers for macOS, Windows, Linux, and Docker and model families from lightweight laya to more accurate decider variants. HN reactions centered on accuracy versus convenience: several testers said laya handles simple classifications but trails Jev on complex semantics and longer context, while others argued quick fine-tuning on private data closes the gap at far lower cost than renting large-model inference, making these models ideal for pushing lightweight classification down from the cloud. Discussion: Hacker News thread.

What about Rails when coding is increasingly generated?#

Jared Norman’s essay What About Rails? recaps DHH’s opening keynote at Rails World 2026: the Rails creator said he has retired as a professional programmer and now calls himself a maker, argues English is the best programming language and that reading generated code need not be the norm, and cites writing about 150,000 lines in August with LLMs versus tens of thousands per year before. In that telling, the next Hey front ends go native on each platform and the backend moves to Rust because models handle it well and humans skim less. The proposed norm is services exposing CLIs for agents. The essay worries this leaves Rails’ core promise to small teams less clear. Commenters split: optimists said handing repetitive code to models lets engineers focus on product and interfaces; skeptics warned that not reading code stores up maintenance and liability, and asked the framework to publish clearer norms and guardrails for an AI-heavy workflow. Discussion: Hacker News thread.

Society and Culture#

Factorio you can touch, from assembly lines to STL files#

Friday Facts #447 turns the factory game into a tabletop system. After a 2024 playtest with Prusa Research and hands-on 3D printing at the event, the team spent spare time refining a grid built around belts — the game’s core — with designers Jaroslaw and Fearghall building inserters, spawners, chests, and the engineer character. The release narrows to the early game: 15 sets, 65 models, 247 STL files, with high-precision push-fit and high-clearance glue-up options to suit different printers. The post shows printed and painted builds and positions the collection as a way to recreate in-game assembly lines on a desk. Readers responded with cheerful shop talk about filament, tolerances, and paint, and praised the choice to do the early game well with clearly organized files rather than chasing breadth. Discussion: Hacker News thread.

Should babywearing carriers be allowed on planes?#

Jeff Kaufman’s post Allow babywearing carriers on planes revisits a knot in FAA rules: lap infants under two are allowed, in part because forcing a ticket purchase would push some families to drive — and driving is less safe than flying — but carriers must be removed for taxi, takeoff, and landing and the baby held in arms. Walking back through the 1994 FAA tests and the 1995-1996 rulemaking, he notes the studies covered booster and harness-style restraints but did not benchmark the most common alternative — arm-held — and rested on the phrase that carriers may put children in a worse situation than allowable alternatives. He cites differing approaches in the UK and Europe and a 2024 U.S. bill that did not advance, arguing that without modern carrier testing the default ban lacks evidence and calling for controlled comparisons across turbulence, hard braking, and evacuation scenarios. The thread weighed probabilities carefully. One camp warned about projectiles if rigid seats are loose and about slower egress when straps are slow to release; the other argued that banning without data penalizes a device that keeps hands free and may retain the child better than arms in turbulence, urging scenario-based evidence. Discussion: Hacker News thread.

The Test, or why strange boasts can be a loyalty check#

In The Test, blogger tante starts with Jensen Huang telling Ezra Klein he does not know his own address or ZIP code and does not think basic skills matter, then links similar recent remarks from other tech leaders. The argument is that such throwaway boasts function as a test: if interviewers let the absurd pass, it normalizes detachment and rewards deference. Quoting William Gibson on how great wealth can make people seem no longer human, the post says money and software both create distance from the friction of ordinary life — when someone else handles daily details, it is easy to discount the value of skills and institutions. The author urges listeners and journalists to pull such claims back to ground level and name their limits and costs, so audiences can fail the test on purpose and keep public conversation anchored. HN reactions circled the tool-versus-ability boundary. Some said forgetting long division or a ZIP code is simply what tools are for; more argued children still need hands-on practice to build number sense and judgment, and that the media’s lack of pushback turns eccentricity into vision. Discussion: Hacker News thread.

Closing#

From replacing a monolithic workplace suite with inspectable blocks to asking whether contracts, tax bases, and everyday rules are built on verifiable evidence, today kept returning to the same word: control. Not control for its own sake, but the ability to audit the stack, to name where responsibility sits, and to keep paying for maintenance after the announcement. The smaller stories point the same way — chip-level speed that any Go program can reach, judgments that stay on your machine, a framework negotiating its next decade, a game that leaves the screen — and the human stories close the loop by asking whether our tools make us more aware of the world or more insulated from it.

See you next time.