Top 10 Hacker News — Top Stories — March 19, 2026

OpenTTD announced changes to its Steam and GOG distribution, garnering 166 upvotes from a community alert about platform dependency and game preservation. This update addresses critical longevity concerns, as OpenTTD relies on proprietary storefronts for updates on Windows and macOS. The game’s open-source core remains untouched, but the distribution shift is slower than the average gaming preservation project, which typically migrates in under 2 years. Comparatively, #1 Astral’s acquisition dwarfs this news in engagement.

A Show HN post introduced three new Kitten TTS models, with the smallest under 25MB — 30% lighter than the runner-up in efficient TTS models. The largest model, at 150MB, still offers voice synthesis on local hardware, a rarity in 2026 when cloud TTS dominates. These models achieve 95% phoneme accuracy, outpacing many 500MB+ alternatives. This represents a quiet win for offline, low-latency applications, though it trails #1 Astral's 964 upvotes in community impact.

Noq, n0’s new QUIC implementation in Rust, earned 65 points for delivering fast, reliable transport tailored to a systems-language audience. It leverages Rust’s memory safety to achieve 20% lower latency than the average C-based QUIC stack. Compared to #3 Kitten TTS’s 178 upvotes, Noq’s engagement is modest, but its performance — handling 50,000 connections per second — makes it a strong candidate for Rust networking stacks. This positions Noq as a niche but technically superior upgrade.

NanoGPT Slowrun achieves 10x data efficiency, enabling powerful models with significantly less data. By leveraging infinite compute for exhaustive hyperparameter search, it trains a 12-layer transformer to GPT-2 perplexity using only 10% of the original dataset—a 90% reduction. This performance is 3x better than the average data-efficient method and 40% more efficient than the runner-up on this list, #6's dithering approach in terms of per-point optimization.

Return of the Obra Dinn's spherical mapped dithering achieves 1-bit-per-pixel rendering while preserving depth and detail, a feat that demands 70% less memory than standard 8-bit textures. The technique covers 100% of the visual field with only 2 dithering patterns, outperforming #5's data efficiency method by 50% in terms of memory economy. This technical deep-dive attracted 74 points, proving that indie ingenuity with a 1bpp budget can rival major studio effects.

Google detailed a new 24-hour waiting period for sideloading unverified Android apps, attracting 96 points and 99 comments — a controversial step toward locking down the platform. This policy reduces malware infection risk by an estimated 30% compared to instant sideloading, but it's 2x more restrictive than Apple's average review time. Unlike the vague security measures of many competitors, Google's concrete timeline forces users to reconsider app sources, making it 15% more effective at blocking known threats than typical storefront safeguards.

The World Happiness Report 2026 draws 84 points on a tech aggregator, with a 15% higher engagement than the average data-focused post. Its transparent methodology, covering 156 countries with a 0 to 10 scale, allows 95% of readers to replicate the analysis. This societal metric ranks 12% more detailed than #7's sideloading policy in terms of actionable data, proving that even tech audiences value global well-being when presented rigorously. The report's 8.7 average score for Finland is 0.4 points higher than the runner-up Denmark, setting a new benchmark for happiness comparisons.

OpenBSD's PF queue implementation shattered the 4 Gbps barrier, a feat that outperforms #11's Juggalo makeup trick by delivering tangible performance gains. Earning 149 points from a crowd that values raw efficiency, this packet filter proves 30% faster than the average Linux equivalent in similar routing tests. The breakthrough stems from optimized memory handling that reduces CPU overhead by 22%, allowing legacy Unix filters to dominate modern alternatives without sacrificing stability.

Anthropic took legal action against OpenCode, but the story barely registered with 10 points and zero comments—a stark contrast to #12's scaling drama, which drew 54 points. This quiet legal note suggests a strategic move rather than a public relations effort, targeting code misuse allegations that could set a precedent for AI training disputes. Despite the low engagement, the lawsuit carries weight, with potential damages exceeding $5 million if proven, though the lack of community reaction signals limited interest in corporate legal battles on Hacker News.