News for hackers. Stuff that matters.

joe-excom

Re:HTML is Alive – an exhortation to remove its ventilator on Thursday October 22, 2026 @12:21PM (#50197514)
Hello HN - I'm Joe, the author of this post. Nucleus is not a weekend-vibecoded project. I took a 3.5 year hiatus from consulting to become a rancher and have been working on this project since then.

I appreciate any feedback - even criticism - on the architecture & Nucleus implementation.

Man, it's hard to promote here. I just posted one of my own. Little traction. Looks like we have some similarities in the goals of our projects. Wonder if they can synergize somehow.
Hi HN. I'm Joe, the solo developer behind this project. I left my consulting career 3.5 years ago; I became a farmer/rancher and have been diligently spending my remaining free time on this project.

The Nucleus stack is a few decoupled web technologies that work together as an implementation of the "ASO" architecture (stands for Adapter, State, Orchestrator). Practically, the goal of Nucleus is sophisticated, declarative, simple web apps with low-to-zero JS and no build process.

Architecturally, the "Orchestrator" is the primary innovation here. The original motivation: the web has never had an easy, native way to map data to the document. Every web framework primarily exists to fill this gap; most achieve it by keeping a copy of app state in JS memory and treating the DOM as a compile target. Yet CSS has quietly held templating-like abilities for a long time:

    span::before { content: attr(data-text); }
The "Orchestrator", dubbed "Quark", takes this idea seriously. It's a CSS derivative whose properties mutate the document, rather than style it. It can set attributes, iterate lists, stamp templates, listen for events, dispatch commands, run View Transitions, and more.

"Adapters" are certain elements (native or custom elements) that each bridge the "State" to one external protocol, such as fetch, form, history, WebAuthn, user gestures. They write only their own attributes.

"State" is simply the live DOM.

Two demos to take a look at; both have zero lines of app JS and no build process:

- The Todo demo on Nucleus' landing page is a fully RESTful app in about 30 lines of HTML, 20 lines of Quark.

- The Wrenfield PWA demo is a fictional, full-featured ecommerce site with polished features like SSG, SPA, hydration, web AR/3d models, WebAuthn, View Transitions, touch gestures. Try it on mobile to get a native-like experience. https://wrenfield.excom.dev

The Nucleus site itself is also dogfooding the stack. Opening up your inspector will show a highly declarative document, which is the live State of the app. Other examples are in the sidebar. The Nucleus stack is currently being used in production by a few companies. Open source versions of their apps are coming soon, as demos.

The stack differs from things you're probably aware of:

- HTMX - server owns logic and swaps fragments. Quark owns orchestration on the client, with the document as the only state.

- Lit and other web component frameworks - Custom elements are also part of this stack, and there can potentially be interop. But the ASO architecture advises elements to never render their own opinionated children or business logic. That is left to you.

- Datastar / Alpine - JS expressions or signals inside attrs. Quark lives in a separate, decoupled sheet and has no expressions with side effects. It does not call anything except your named pure functions.

- Blackboard pattern - architecturally, this is the closest relative. The site has an "Advanced -> Prior Art" page which discusses this.

The Nucleus stack is in beta. Here are the edges:

- Quark: unlike CSS, rules will not revert when they stop matching. You need to write the inverse, if needed. This is by design. Regardless, I'm currently exploring the architectural feasibility of rule reversion.

- Quark: certain pseudoclasses like :focus, :checked, :invalid are not observed.

- Quark and the elements rely on some fairly new browser APIs (like the Command API, Baseline 2025). Polyfills will be included in the first stable release.

- Will not play with React and certain other frameworks that lock the DOM to their own internal state. You will need a Shadow DOM as a boundary between them and the Nucleus stack.

Depending on the bundle/loading strategy you choose, the whole stack is somewhere between 1kb and a maximum of 70kb (brotli). This includes all out-of-the-box elements that provide generic functionalities like data fetching, lazy loading, and UI things like drawers, tabs, etc.

I am very keen on hearing feedback on this project. Particularly on the ASO architecture and the Quark language.

Nucleus will remain in beta for a few more weeks, as features are optimized and feedback is incorporated.

Docs: https://nucleus.excom.dev

Repo: https://github.com/excom-dev/nucleus

Architecture: https://nucleus.excom.dev/docs/adapter_state_orchestrator

You could replace "math" with "programming" in your comment for a different perspective.
Re:Rust Port of TypeScript (Tsc) on Wednesday October 07, 2026 @11:20PM (#50001562)
It teaches us the Rust community continues to be relentless in their efforts to rewrite virtually everything in Rust.
Re:Why don't more developers "use the platform"? on Tuesday October 06, 2026 @04:30AM (#49975777)
Truthfully, one primary reason: the web natively lacks the ability to declaratively map data to HTML. Virtually all web frameworks exist primarily to fill this gap. Non-native implementations of other features outflow from that.
Re:Uber Wait Times Rise 19% as Prices Climb 53%: Study on Thursday October 01, 2026 @12:46AM (#49917844)
In the last few years, I have not had a single Uber ride from the airport to my house (15 mins) cost less than $60. Sometimes it's as high as $120. I recently went back to carpooling and being social.

Uber has over 30k employees... why? The service quality has not increased in a decade.

Ride apps are on an enshittification speedrun, worse than virtually any other company or service out there.

Completely in agreement, I had the exact same thought. Reminds me from a few years ago where Sam Altman went in front of Congress to scaremonger the country over AI. Purely a thinly-veiled plead for regulatory capture.
I see you were turned away from submitting a "Show HN" too. I might have to do the same as you for my project.
Re:When did Google get so f-ing weird? on Monday September 28, 2026 @02:18AM (#49874277)
I have similar experiences. Google needs to realize their search variant of Gemini needs guard rails... in order to actually serve the purpose of search. Gemini also seems to hallucinate the most frequently out of any of the big LLMs. Doubly bad when it's repackaged as a search assistant.