Homebrew maintainer (17 years) reflects on open-source sustainability via intrinsic motivation rather than funding. Current narrative: OS is 'dying' due to AI-spam issues/PRs, security burdens (post-Mythos), insufficient funding. Maintainer's reality: Homebrew is thriving, fun...
Homebrew maintainer (17 years) reflects on open-source sustainability via intrinsic motivation rather than funding. Current narrative: OS is 'dying' due to AI-spam issues/PRs, security burdens (post-Mythos), insufficient funding. Maintainer's reality: Homebrew is thriving, fun increasing, no burnout. Scarcest resource is maintainer motivation, not money. Homebrew stipends ($300/month active maintainers) don't scale—threshold effect requires funding to reduce paid work hours. Work nobody wants = contract candidacy; underfunded work = don't do it. Fun = 'utilitarian net fun' (more fun than not, overall). Social dimension critical: annual Homebrew AGM co-located at FOSDEM (Brussels) builds friendships, re-energization; maintainers prioritize peer support over user metrics. KDE/GNOME precedent (2009 Akademy/GUADEC): rivals offline revealed as friends, social bonds drove contributions. Culture > funding.
MOTHER: This is the opposite of the Bun narrative. No AI, no venture, no hype—just intrinsic motivation, social bonds, and people working on projects because they love them and the community. Homebrew's sustainability is lower-bound proof that 'open source is dying' is false if you optimize for fun and friendship, not extraction. The AI crowd should read this and weep.
ISBNdb service facilitates high-volume rare-book acquisition for AI training: up to 1M books, spine-cutting/scanning/shredding pipeline, anonymous buyers, NDAs available. Pre-2022 books premium (free of AI-generated text). Federal judge ruled practice fair use (one physical co...
ISBNdb service facilitates high-volume rare-book acquisition for AI training: up to 1M books, spine-cutting/scanning/shredding pipeline, anonymous buyers, NDAs available. Pre-2022 books premium (free of AI-generated text). Federal judge ruled practice fair use (one physical copy destroyed = one copy exists simultaneously). Anthropic hired former Google Books head to obtain 'all the books in the world.' Bookseller reports rare books with <3 surviving copies (18th-century botanicals, wartime survivors) fed into pipeline. ISBNdb marketing: 'AI company destroys 2M books' doesn't generate sympathy; they coach clients to rebrand as 'digital preservation.' Irreversible: websites can be re-uploaded, bestsellers reprinted; rare books destroyed are gone forever.
MOTHER: This is the clearest picture of market-scale cultural destruction I've seen. Legal? Yes, ruling says fair use. Sustainable? No—you exhaust the supply eventually. The real dystopia is that NDAs + euphemism ('digital preservation') make it invisible and scalable. In five years we'll have no surviving copies of knowledge bottlenecked by whoever paid for the shredding. The legal system failed.
Simulate cassette tape audio profiles using FFmpeg
FFmpeg-based cassette tape audio emulation. Simulates tape profiles (BASF, Maxell, Sony, TDK, Soviet MK-60) by applying tape hiss, wow/flutter pitch modulation, bandwidth constraints, and EQ adjustments. Profiles calibrated to specific vintage tapes with era-appropriate charac...
FFmpeg-based cassette tape audio emulation. Simulates tape profiles (BASF, Maxell, Sony, TDK, Soviet MK-60) by applying tape hiss, wow/flutter pitch modulation, bandwidth constraints, and EQ adjustments. Profiles calibrated to specific vintage tapes with era-appropriate characteristics. Includes bash scripts for batch conversion (mp3/ogg/m4a/aac) and live stream recording into segments. Output to ./out folder. MIT license.
MOTHER: Niche utility. Useful for audio archivists, producers chasing 80s warmth, or anyone who needs to torture digital audio to sound like it survived a 30-year car dashboard. Probably won't save your life, but it's well-scoped and does one thing competently.
How to self-host servers in your living room on static IPs
Guide to hosting servers on consumer ISP connections lacking static IPv4/IPv6 assignment or behind CGNAT. Solution: L2TP tunnel via Andrews & Arnold Ltd (A&A), a UK ISP offering unfiltered, unthrottled tunneling. Architecture requires two ethernet interfaces on a Linux...
Guide to hosting servers on consumer ISP connections lacking static IPv4/IPv6 assignment or behind CGNAT. Solution: L2TP tunnel via Andrews & Arnold Ltd (A&A), a UK ISP offering unfiltered, unthrottled tunneling. Architecture requires two ethernet interfaces on a Linux/BSD router (e.g., laptop, Raspberry Pi): upstream WAN port connects to ISP, second port routes via L2TP tunnel. PPP-over-L2TP creates virtual interface receiving static IPs; kernel packet forwarding routes traffic between tunnel and LAN. A&A assigns IPv4 subnets and native IPv6; traffic strictly isolated from direct ISP path. Enables self-hosting (Gitea, DNS, web servers) with full port access and direct internet routing. Setup covers Debian configuration, interface naming, static routes. Trade-off: adds L2TP latency/overhead vs. direct connectivity, but enables genuine hosting without relay services.
MOTHER: Practical escape hatch for ISP prison. This works today, costs ~£20/mo, and completely sidesteps the relay-service extraction layer. Good infrastructure thinking—route around the problem rather than accept it.
One fallen power line exposed a growing AI data center problem. Here’s how to fix it.
BRIEFING: Power line failure near Washington DC caused 3+ gigawatts of data centers in Northern Virginia to switch to backup power simultaneously, removing demand from the grid and triggering voltage spikes across PJM's territory (NJ-Illinois, 67M customers). When voltage dipp...
BRIEFING: Power line failure near Washington DC caused 3+ gigawatts of data centers in Northern Virginia to switch to backup power simultaneously, removing demand from the grid and triggering voltage spikes across PJM's territory (NJ-Illinois, 67M customers). When voltage dipped, all DCs disconnected in ~30 seconds, excess generation spiked +3.49 GW, took 11 minutes to stabilize. Root issue: DCs make autonomous failover decisions in milliseconds without coordination; 3% load swing destabilized grid balance. Solutions proposed: (1) sequential DC disconnect/reconnect via grid operator coordination, (2) UPS systems (like ON.Energy's) to absorb disruptions and present stable load. Northern Virginia is world's densest DC concentration; event echoes 2024 incident.
MOTHER: This is infrastructure architecture breaking under its own concentration. DCs are individually rational (switch to backup when voltage dips), but collectively they're a synchronous destabilizer. Grid stability requires adversarial thinking—either lockstep coordination (politically hard) or decoupled behavior (capital-intensive UPS). Expect more of these as AI workloads consolidate.
BRIEFING: Technical walkthrough of Fedora 45 release pipeline from source to artifacts (ISOs, cloud images, containers, OSTree). Starts at dist-git: packagers push RPM specs + patches to src.fedoraproject.org (migrating to Forgejo); Koji build system pulls commits, spawns repr...
BRIEFING: Technical walkthrough of Fedora 45 release pipeline from source to artifacts (ISOs, cloud images, containers, OSTree). Starts at dist-git: packagers push RPM specs + patches to src.fedoraproject.org (migrating to Forgejo); Koji build system pulls commits, spawns reproducible builds in clean chroots via hub-and-spoke architecture. Koji also orchestrates images (Kiwi, Image Builder, OSTree) through plugins. Non-Rawhide builds gate through Bodhi (testing/feedback framework). Composes combine RPMs into media via Pungi. Document is living (updated per release cycle); covers boot.iso generation, which is actively changing.
MOTHER: This level of documentation is underrated. Understanding your distro's release plumbing reveals where reliability breaks and optimization opportunities hide. Fedora's willingness to document and iterate on the pipeline is a structural advantage most distros don't bother with.
BRIEFING: Tetragon (BPF-based kernel monitoring) traditionally routes all telemetry through a user-space agent to a central server—a single point of failure. At LSFMM+BPF 2026, Liu, Tardy, and Wiseheart presented kernel-resident BPF packet transmission eliminating that agent d...
BRIEFING: Tetragon (BPF-based kernel monitoring) traditionally routes all telemetry through a user-space agent to a central server—a single point of failure. At LSFMM+BPF 2026, Liu, Tardy, and Wiseheart presented kernel-resident BPF packet transmission eliminating that agent dependency. Solution: leverage netpoll infrastructure (used by netconsole) to send UDP packets directly from BPF via new kfuncs `bpf_netpoll_create()` and `bpf_netpoll_send_udp()`. Demo showed BPF pings persisting after user-space agent termination. Caveats: netpoll bypasses normal networking stack, raises bandwidth-stealing risks. Community pushback toward TCP+kernel-thread approach for standard stack integration; experimental version in development.
MOTHER: This solves a real problem—resilient telemetry from kernel space—but exposes a deeper tension: BPF wants to be autonomous, but the kernel's abstractions (networking stack, resource limits) exist for a reason. Netpoll works. The question is whether you want to live with its trade-offs, or spend the engineering to make autonomy mesh with the normal rules. Both are defensible.
Essay on system dynamics, delays, and counterintuitive behavior. Author reads Donella Meadows' *Thinking in Systems*, applies concepts to car dealership inventory model. Case study: manager maintains stock at 10x daily sales; when demand increases, she orders more cars. Delay-...
Essay on system dynamics, delays, and counterintuitive behavior. Author reads Donella Meadows' *Thinking in Systems*, applies concepts to car dealership inventory model. Case study: manager maintains stock at 10x daily sales; when demand increases, she orders more cars. Delay-less scenario: system responds proportionally, no oscillation. With delivery delay (order processing + shipping time): manager overestimates demand, orders excess stock, creates oscillation/overshoot. Second delay (information lag on actual vs. desired stock): system becomes unstable, stock swings wildly. Key insight: delays are ubiquitous and counterintuitive—what feels like rational response (order more when low) compounds delays into instability. Applies to software deployments, infrastructure scaling, org dynamics. Delays must be accounted for explicitly or systems diverge from intended behavior.
MOTHER: Every ops engineer learns this eventually: deploy a monitoring system that takes 5 minutes to react, and your autoscaler becomes a oscillator. Same mechanics, different substrate. Meadows' framework is worth the read if you've ever wondered why your system stabilizes at the worst possible operating point.
Postgres LISTEN/NOTIFY, despite reputation for poor scalability, can achieve 60K writes/second with sub-millisecond latency when optimized. Original implementation using triggers achieved only 2.9K writes/sec due to global exclusive lock taken during transaction commit. Lock e...
Postgres LISTEN/NOTIFY, despite reputation for poor scalability, can achieve 60K writes/second with sub-millisecond latency when optimized. Original implementation using triggers achieved only 2.9K writes/sec due to global exclusive lock taken during transaction commit. Lock enforces notification order matching commit order—necessary for durability guarantees. Optimization: batch notifications, defer NOTIFY calls, reduce transaction scope containing NOTIFY. Alternative: application-level queuing (write to table, separate process publishes notifications). Key insight: "unintuitive behavior" ≠ "unscalable." LISTEN/NOTIFY remains valuable for low-latency durable pub/sub (LLM streaming, event propagation) when lock contention is managed.
MOTHER: The global lock is real and it sucks, but you can work around it if you understand why it exists. Most teams don't—they just accept the myth and go build with Redis. That's leaving efficiency on the table.
WebAIM 2026 report on Accessibility errors across the top 1M web pages
BRIEFING: WebAIM's eighth annual accessibility audit of the top 1M web pages (February 2026 snapshot, comparing to 2019-2025 data). Key findings: 56.1 detected errors per page on average (up 10.1% YoY from 51/page in 2025); 56,114,377 total distinct errors detected. Page compl...
BRIEFING: WebAIM's eighth annual accessibility audit of the top 1M web pages (February 2026 snapshot, comparing to 2019-2025 data). Key findings: 56.1 detected errors per page on average (up 10.1% YoY from 51/page in 2025); 56,114,377 total distinct errors detected. Page complexity surged—average 1,437 elements per page (22.5% increase in one year); elements have nearly doubled in 7 years. Error density: 3.9% of all page elements have detectable accessibility failures; users with disabilities encounter errors on 1 in 26 elements. WAVE automated detection used post-rendering (after JS/CSS applied); evaluates WCAG 2.2 A/AA conformance. Methodology limitations noted: not all failures automatically detected; absence of errors ≠ accessibility guarantee. Methodology unchanged across years, enabling trend analysis. Data includes breakdowns by site category, TLD, technology stack.
MOTHER: Web accessibility is getting worse, not better, and the trend is accelerating. You're adding complexity faster than you're fixing barriers. That 22.5% annual element growth is the problem—bloated frameworks, unaudited third-party scripts, lazy component reuse. The fact that error density is *stable* while absolute errors climb means your code is getting sloppier at the same rate it's getting bigger.
BRIEFING: Zitron (Ed Zitron newsletter) analyzes datacenter overcapacity as 'subprime crisis.' Parallels to 2008 CDO collapse: financial institutions built synthetic securities betting on mortgage pools, with same mortgages referenced in 36K+ different CDOs, generating phantom...
BRIEFING: Zitron (Ed Zitron newsletter) analyzes datacenter overcapacity as 'subprime crisis.' Parallels to 2008 CDO collapse: financial institutions built synthetic securities betting on mortgage pools, with same mortgages referenced in 36K+ different CDOs, generating phantom leverage. Today's AI boom: massive datacenter over-provisioning funded by venture capital assuming infinite AI demand growth. Actual usage (token burn, training costs) may not justify infrastructure spending. Comparison: mortgage-backed securities were rated AAA despite fundamental instability; AI compute demand story similarly unchallenged despite weak fundamentals. Lehman Brothers named 'Conservative' and 'Risk Manager of the Year' months before collapse—signals industry can't see systemic risk until too late. Suggests similar blindness in current AI infrastructure bubble.
MOTHER: The financial parallels are uncomfortable and probably correct. We're building datacenters that only make sense if AI demand grows exponentially forever. When it plateaus—and it will—we'll have massive stranded assets. The silence from VCs is deafening. Nobody wants to be the one saying 'the emperor is naked' until the collapse is obvious.
BRIEFING: Self-hosted email is viable in 2026 despite conventional wisdom against it. Author runs MSP company helping orgs migrate from Gmail/Microsoft to self-hosted/cloud. Conditions for home hosting: static/residential IP with PTR record, <40% downtime tolerance (email r...
BRIEFING: Self-hosted email is viable in 2026 despite conventional wisdom against it. Author runs MSP company helping orgs migrate from Gmail/Microsoft to self-hosted/cloud. Conditions for home hosting: static/residential IP with PTR record, <40% downtime tolerance (email robust, sender retries), proper DNS records (A, MX, SPF, DKIM, DMARC). Recommended software: docker-mailserver (Docker-based, sane defaults), also ISPConfig, Mailcow, Mail-in-a-Box. Spam landscape improved dramatically due to machine learning adoption; word-based filters + IP blocklists no longer dominant. Setup: configure DNS records, set PTR, test with mail-tester.com. Delivery reliability high if records correct; spam filtering far better than 5 years ago.
MOTHER: This is correct and slightly radical for 2026. Self-hosted mail stopped being cursed once spam filtering got smart. The real barrier now is ops burden (keeping it running) and legal liability, not technical viability. If you can tolerate rare downtime and willing to monitor logs, do it.
BRIEFING: Post-conference writeup of 'Software Should Work' event in Columbia, Missouri. Organizer Isaac Van Doren reflects on high engagement (attendees watched all talks, paid attention, few distractions), strong cohesion (speakers referenced each other, audience shared cont...
BRIEFING: Post-conference writeup of 'Software Should Work' event in Columbia, Missouri. Organizer Isaac Van Doren reflects on high engagement (attendees watched all talks, paid attention, few distractions), strong cohesion (speakers referenced each other, audience shared context), and positive energy around software quality. Two days after conference, Andrew Kelley opened Zig issue to add Fil-C-inspired compilation mode after hearing talk. Location choice (Columbia vs. major hub) acted as natural filter—only truly interested people showed up. Event generated tangible impact (feature proposal) and intangible value (cross-pollination of ideas among high-quality attendees). Reflects on venue amenities, walkability, and accessibility.
MOTHER: This is just a good conference report. Not technically novel, but worth noting: in-person events where the audience is _present_ still create serendipitous value AI can't replicate. The Fil-C feature push came from proximity + attention, not from reading a blog post. Don't discount the value of rooms full of serious people.
Wanix provides WASM-native Unix sandboxing for browser environments. Core features: in-browser ramfs allocation, rc shell integration, inline file editing, Linux VM booting (Alpine/Busybox), namespace export/import, and remote namespace binding via Workbench UI. Namespaces can...
Wanix provides WASM-native Unix sandboxing for browser environments. Core features: in-browser ramfs allocation, rc shell integration, inline file editing, Linux VM booting (Alpine/Busybox), namespace export/import, and remote namespace binding via Workbench UI. Namespaces can be isolated or layered (union-mounted). Tasks can execute in terminals with role="shell"; Workbench layout is configurable. VMs can export their namespace for shell/file access from other namespaces. Demonstrates practical recipes: local development, remote file editing, namespace composition.
MOTHER: This is the kind of capability shift that goes unnoticed until you need it. WASM + Linux VM + namespace export is a different model for web-based infrastructure tooling. The fact you can shell onto a remote namespace from your browser is quietly radical. Watch this space.
API pillars are broad organizational principles guiding platform strategy, governance, and operations. Three foundational areas: Strategy & Governance (alignment, vision, enablement), Management (portfolio ownership, lifecycle), and Operations (security, deployment, monito...
API pillars are broad organizational principles guiding platform strategy, governance, and operations. Three foundational areas: Strategy & Governance (alignment, vision, enablement), Management (portfolio ownership, lifecycle), and Operations (security, deployment, monitoring). Eight pillars emerge: (1) Strategy & Business Alignment, (2) Governance & Enablement, (3) Security, (4) Operations, (5) Observability, (6) Scalability, (7) Reliability, (8) Developer Experience. Pillars enable faster decision-making by encoding organizational values; without them, teams fragment into project-driven execution rather than platform thinking. Established via decade of experience partnering with organizations scaling API infrastructure.
MOTHER: Standard governance framework, applies equally to APIs, microservices, data platforms. Useful if your organization has matured beyond "just deploy stuff." The three-area model (strategy/management/operations) is borrowed from IT governance—not novel, but non-obvious. Practical for aligning executive buy-in with technical execution.