HackerNews Digest

98.css is a pure‑CSS design system that styles semantic HTML to mimic the Windows 98 desktop UI. It requires no JavaScript, making it framework‑agnostic; it can be loaded via CDN (e.g., unpkg) or installed with npm. The library provides styled components including: * **Button** – default size 75 × 23 px, raised borders, “default” class for primary actions, visual states for pressed, disabled, and focused. * **Checkbox** and **OptionButton** – require associated `<label>` elements; support `checked` and `disabled` attributes and consistent spacing via `.field‑row`. * **GroupBox** – implemented with `<fieldset>`/`<legend>` for framed control groups. * **TextBox** – single‑line `<input type="text">` or multiline `<textarea>`, optional `.field‑row‑stacked` for vertical label placement. * **Slider** – `<input type="range">` with optional `.has-box-indicator` and vertical orientation via `.is-vertical`. * **Dropdown**, **TreeView**, **Tabs**, **TableView**, **Progress Indicator**, and **Status Bar** – each rendered with standard HTML elements plus specific CSS classes (e.g., `.tree‑view`, `.progress‑indicator`, `.segmented`). Window structures combine `.window`, `.title‑bar`, `.window‑body`, and optional `.status‑bar`. Accessibility is emphasized through proper use of `aria-label` and label associations. The project is MIT‑licensed, with source, releases, and issue tracking on GitHub.

None

The page is identified by the title “Visual 6502 in JavaScript,” indicating that it concerns a JavaScript‑based implementation of the Visual 6502 project. Visual 6502 is an interactive, visual simulator of the MOS 6502 microprocessor, originally created to illustrate the internal operation of the CPU cycle‑by‑cycle. By re‑creating it in JavaScript, the project enables the emulator to run directly in web browsers, providing real‑time visualizations of registers, memory accesses, instruction decoding, and bus activity without requiring native code. This approach makes the 6502 architecture accessible for education, debugging, and hobbyist experimentation, leveraging standard web technologies (HTML5 canvas or SVG) to render the CPU’s internal state dynamically as programs execute.

None

Plex dominated home‑media streaming for a decade, offering self‑hosted organization and wide‑device support. Recent years show a strategic shift toward ad‑supported FAST channels and licensed studio content, pushing external media to the default home screen and introducing subscription friction such as Remote Watch Passes. Centralized authentication via Plex.tv creates outage risks and data‑collection concerns; Jellyfin authenticates locally, avoiding cloud dependence. Plex has improved features like mobile sync but removed others (e.g., Watch Together) while emphasizing discoverability and social feeds. Jellyfin, a fully open‑source fork of Emby, provides local accounts, zero tracking, and free GPU transcoding, but its client ecosystem is fragmented across volunteer projects. In July 2026 Jellyfin faced a leadership exodus, highlighting open‑source maintainer burnout and AI‑generated PR overload. Plex’s pricing escalated from a $75‑$90 lifetime pass (pre‑2021 sales) to $749.99 in July 2026, adding a $249.99 five‑year tier and $6.99 monthly option. The author recommends Jellyfin for new self‑hosters seeking full control, while retaining Plex for existing users who value its mature client support and features.

The guide explains the operation of a beam engine from basic steam principles to full‑scale mill power. Water heated in a boiler generates steam that expands ~1,700 × its liquid volume; the resulting pressure pushes a piston whose force equals steam pressure times piston area. Early engines, limited by weak low‑pressure boilers, created a vacuum by condensing steam, allowing atmospheric pressure to drive the piston downward (Newcomen’s engine). Watt improved efficiency by adding a separate cold condenser, keeping the cylinder hot and reducing coal use by two‑thirds, and by making the engine double‑acting so steam acted on both piston strokes. A slide valve, driven by an eccentric, timed steam admission and exhaust. Reciprocating motion was converted to rotation via a crank and flywheel; the parallel‑motion linkage kept the piston’s motion linear. A small pump fed the boiler continuously. A centrifugal governor automatically regulated steam flow to maintain constant speed. Subsequent advances—higher‑pressure boilers, compound expansion, Corliss valves, and turbines—raised thermal efficiency from <1 % (Newcomen) to >40 % in modern plants.

None

The Numbers, a long‑standing, free database of box‑office, budget and personnel data for ≈78 000 films, went offline on 5 Mar 2026 after its servers were overwhelmed by AI‑generated traffic. Founder Bruce Nash explained that human visits now represent only ~10 % of requests; the rest come from search‑engine crawlers, large‑scale LLM‑training scrapers, and “agentic” AI bots that repeatedly scrape pages and probe for back‑doors. The surge in automated traffic in 2024–2025 forced the team to spend ~90 % of their effort on keeping the legacy site (≈160 000 files) alive, ultimately abandoning the old code and launching a skeletal version on new infrastructure on 13 Mar. The incident illustrates a broader trend: AI tools have lowered the barrier to massive web‑scraping and cyber‑intrusion, causing DDoS‑like loads on sites such as Read the Docs, iFixit, GNOME and Wikipedia. Reports from Anthropic and the emergence of AI‑driven penetration testers confirm that sophisticated attacks can now be executed with minimal human input. Cloudflare’s recent pay‑per‑crawl/pay‑per‑use policies aim to monetize AI access, but the fundamental assumption that most traffic is human is now obsolete. The Numbers’ core revenue—bulk data sales and reporting—remained intact, highlighting the vulnerability of public‑facing sites that rely on legacy architectures in an AI‑intensive web ecosystem.

- Freeze‑casting originated in 1954 (Maxwell et al.) for dense ceramics; modern porous‑structure use began with Fukasawa et al. (2001) and has expanded to hundreds of papers. - A suspension (typically water‑based) is frozen; ice crystals template the pore network, then sublimated, leaving a green ceramic preform. - Three solidification regimes: planar front (no macroporosity, < 1 µm s⁻¹), lamellar/cellular templating (targeted porous structures), and particle engulfment (high velocities, large particles, high solid loading). Critical engulfment velocity \(v_c\propto 1/R\) depends on interfacial energies, film thickness, viscosity, and particle size. - Freeze‑cast microstructures exhibit three zones: Initial Zone (no macropores), Transition Zone (mixed crystal orientations, r‑crystals → z‑crystals), and Steady‑State Zone (aligned lamellae). Pore wavelength \(\lambda =A v^{-n}\) links solidification speed to wall‑pore size. - Pore size and shape are tuned via solidification velocity, solids loading, additives (e.g., NaCl, glycerol), and static vs. dynamic temperature gradients; dynamic profiles can maintain constant \(v\) and uniform pore thickness. - Mechanical response follows anisotropic cellular‑solid models: Young’s modulus scales with relative density squared, compressive strength with relative density, and failure is governed by lamella buckling. - Applications exploit aligned porosity for biomedical scaffolds, thermal insulation, high‑performance wicks in loop heat pipes, and porous metal foams for chemical‑loop combustion and steam‑iron hydrogen production.

Software factories that rely on AI coding agents promise speed but often degrade code quality. Early hype (e.g., StrongDM “lights‑off” factory, OpenAI’s Symphony) suggests humans can be removed from code review, yet real‑world data show rising pull‑request comments, incidents, and bugs per developer. The bottleneck shifts from implementation to review; automated linters and adversarial review cannot replace human oversight of maintainability. Models excel at one‑off tasks but struggle to preserve long‑term code health because benchmarks focus on test‑pass rates, not architectural soundness. Reinforcement learning within the tool harness (e.g., Claude Code) improves tool use, but without penalties for design erosion models still produce “slop”—over‑use of try/catch, weak type handling, and fragile patches. Consequently, fully “lights‑off” factories fail: hidden defects force engineers back into the codebase, causing downtime and rewrites. Sustainable AI‑assisted development requires human steering, robust testing pipelines, and new benchmarks that measure maintainability, not just test success.