Formalizing the Tax Code, Catching Starships, and a Hidden Ocean World

Welcome to ReadFlow Weekly. It's the week of November 7th to 14th, 2026, and we're opening with a story about a very different kind of bug hunt. This one isn't in your codebase. It's in the United States tax code. A team is now attempting to formalize the entire Internal Revenue Code using a proof assistant called Lean. The goal is to turn thousands of pages of legalese into a mathematically verifiable system, to hunt for unknown loopholes and contradictions. It's a fascinating collision of law and computer science, a project that asks if you can point a proof checker at a statute and find the bugs. You can read about the ambitious effort to formalize the U.S. tax code.

And that story, in a way, is about building a perfect, abstract system. But this week also delivered a stark reminder that our most critical digital systems are anything but abstract. They are intensely, vulnerably physical. A major piece this week argues that we've spent a decade convincing ourselves the cloud is a magical, weightless place. It is not. It is a building with walls, cooling pipes, and power cables. And sometimes, as recent AWS outages have shown, those buildings get too hot. The piece is a potent call to remember that the cloud is just someone else's computer, one that can, quite literally, burn.

So here we are, caught between two powerful ideas. On one side, the drive to formalize the messy complexities of human law into pure, logical code. On the other, the sobering reality that our digital world rests on a foundation of physical hardware that can and does fail. This tension between our clean abstractions and their messy, tangible foundations is a thread we'll see running through much of this week's news.

From the high-stakes engineering of catching rockets to the economic fallout from corporate scandal, and the hidden signals in our brains that whisper of future illness, this is a week about ambition, cost, and the gaps between the worlds we build and the world as it is. It's all ahead. Welcome to ReadFlow Weekly.

Welcome to ReadFlow Weekly. I'm your host. As we pick up from that opening look at the formalization of the tax code, it's clear this week is about the tension between our grandest ambitions and their very tangible foundations. We're going to see that theme play out across every corner of the show.

In tech and AI, we'll move from that mathematical quest for loopholes in the law to a stark reminder that the cloud, for all its abstraction, is a physical place that can and does fail. We'll also look at the growing energy appetite of AI, particularly for non-English languages, and the engineering feats required to serve frontier models across multiple GPUs. And we'll examine a provocative argument gaining traction: that software is never neutral, and that code itself is becoming politicized.

Our business coverage is dominated by a single, unraveling story: the fallout at Nvidia. The chipmaker's market cap has dipped below four trillion dollars as new allegations surface that the company engineered GPU scarcity, adding fuel to the fire of its leadership crisis. We'll also check in on a high-profile humanoid robot startup that missed payroll amid CEO controversy, and we'll balance the scales of economic ledger with a massive study on the value of COVID vaccines and a record high for Berkshire Hathaway.

Science takes us from the near-miss of engineering history, as SpaceX narrowly failed to pull off a historic double booster catch, to discoveries at the very edges of our knowledge. The Euclid telescope has released a cosmic map of nearly a billion galaxies, while back on Earth, scientists have found hidden brain signals that can predict Alzheimer's years in advance. We'll also see forensic evidence of ancient royal retainer sacrifice in Egypt and the capture of seemingly impossible quantum jumps in electrons.

The thread running through it all is the cost of ambition itself, and the often-surprising gap between our sleek digital abstractions and the physical, messy, and profoundly human reality they're built upon. We've got a deep and fascinating episode ahead, so let's get started.

Welcome back. As we pick up from that opening look at the formalization of the tax code, it's clear this week is about the tension between our grandest ambitions and their very tangible foundations. We're going to see that theme play out across every corner of the show. In tech and AI, we'll move from that mathematical quest for loopholes in the law to a stark reminder that the cloud is just a building that can get too hot, and then to the soaring energy bills and hardware demands of the AI frontier. It's a week where the tools we build, and the politics baked into them, come into sharp focus.

Let's start with that formal verification project, because it's a beautiful and slightly mad idea. A team at Lexifina is now attempting to formalize the entire United States Internal Revenue Code using a proof assistant called Lean. This isn't about writing an interpreter for tax software. This is about translating thousands of pages of dense legalese, with all its conditional clauses and cross-references, into a set of mathematical statements that a computer can logically verify. They've already proven the concept by using Lean to identify known tax loopholes. Now they're going for the whole thing, hunting for unknown bugs in the law itself.

Why does this matter? It matters because law is code. It's a set of instructions for a society to execute. And like any complex codebase written over decades by thousands of different authors, it's riddled with inconsistencies, undefined edge cases, and unintended behaviors we call loopholes. Formal verification is the gold standard in critical software engineering, used for microchips and cryptographic protocols. Applying it to statute is an attempt to bring that same rigor to governance. Who wins if this works? In the long term, citizens and governments win with clearer, fairer laws. In the near term, tax professionals and software companies that can navigate this new, precise map of the tax landscape win. What to watch is whether this project can scale. The U.S. tax code is a monster of complexity. Success here could pave the way for formalizing other complex legal systems, from financial regulations to environmental statutes, turning legal analysis from an art of interpretation into a science of verification.

But while some are trying to elevate code to the purity of mathematics, we got a visceral reminder this week that most of our digital world runs on infrastructure that is profoundly, mundanely physical. A pointed commentary titled The Cloud Is Just Someone Else's Computer , They Let It Burn drove this home, reflecting on recent major AWS outages. The cloud is not a magical, weightless place. It is a building. It has walls, cooling pipes, and power cables. And sometimes, a single data center getting too hot can cascade into a global service disruption. This is the physical limit of our abstraction.

This matters because our entire economy now assumes the cloud is a utility as reliable as electricity. But its foundations are more fragile and concentrated than we often admit. When one provider's building has a problem, huge swaths of the internet go dark. Who loses? Every business and user who discovered their mission-critical application was, in fact, critical to a single physical location. Who wins? Companies offering multi-cloud or hybrid-cloud strategies, and perhaps those arguing for a renewed look at decentralized, physically resilient network designs. Watch for how this narrative influences enterprise contracts and regulatory discussions about cloud concentration risk. The dream of ephemeral, serverless computing crashes into the reality of HVAC systems and diesel generators.

And speaking of physical limits, let's talk about the engine of the current tech boom: artificial intelligence. Two stories this week bookend the AI challenge, one about the cost of ambition, the other about the engineering required to fulfill it. First, the cost. A study from the University of Michigan found that generative AI models use significantly more energy when processing non-English text. The reason is linguistic complexity and the structure of the training data. For many large language models, English is the primary, clean, well-organized training corpus. Other languages, with different scripts, grammatical structures, or less curated data, force the model to work harder, driving up computational cost and energy consumption per token.

This is a massive deal. It means the global expansion of AI services carries a much higher environmental and economic price tag than we anticipated if we want it to be truly multilingual. It risks creating a two-tier AI world: efficient, cheap English-language AI, and more expensive, less accessible AI for the majority of the world's population. Who loses? Everyone advocating for equitable, global AI access, and anyone footing the power bill for massive inference farms. Who wins, in a cynical sense? Energy companies and those who own the most efficient, English-first model architectures. Watch for whether this sparks a research race for more linguistically efficient models, or if it simply gets buried as an inconvenient truth in the scramble for capability.

Now, the engineering ambition. On the hardware side, the frontier keeps moving outward. NVIDIA published a technical blog this week detailing how to serve a massive frontier model, GLM-5.3, across two of their DGX Stations. The key phrase is "the largest models no longer fit on one machine." They're using high-speed RDMA connections over ConnectX-8 adapters to make two separate boxes of GPUs act like one giant, unified pool of memory for model serving. This is a glimpse into the near-term future where running a top-tier model isn't about buying a bigger GPU, it's about networking an entire cluster of them seamlessly.

This matters because it's the infrastructure answer to the model size problem. As long as the race is for more parameters, we need these kinds of distributed systems. It's a win for NVIDIA's ecosystem, obviously, locking in the need for their high-end networking tech alongside their GPUs. But it's also a win for any organization that needs to run these behemoths privately for security or latency reasons. Watch for this pattern to trickle down. The techniques for multi-node, multi-GPU serving developed for these $500,000 setups will eventually become standard for more accessible hardware, defining the architecture of on-premise AI for years to come.

Beneath these massive systems are the tools that developers use every day, and we saw meaningful updates in two foundational areas. First, the C3 programming language released version 0.8.6. The headline features are a revamped library fetch functionality and better support for packaging full-source libraries. It also boasts improved compilation speed and reduced binary sizes. C3 positions itself as a sensible successor to C, and these updates are all about improving the day-to-day ergonomics and efficiency for systems programmers.

Second, and perhaps more groundbreaking, is work from Meta's engineering team on extending the LLDB debugger to debug applications that span CPUs and GPUs. They've built a vendor-neutral accelerator plugin architecture for LLDB. This is a huge deal in a world where more and more computation, especially in AI and high-performance computing, is offloaded to GPUs and other accelerators. Debugging a program where part of the logic runs on a CPU and part on a GPU has been notoriously difficult, often requiring separate, specialized tools. This work aims to bring it all back into a single, coherent debugging session, treating the CPU and GPU as peers in a unified system.

Why do these tooling updates matter? They represent the maturation layer beneath the AI and cloud hype. C3's evolution is about building reliable, performant systems software that the flashier stuff ultimately depends on. The LLDB work is about making the increasingly heterogeneous computing landscape actually manageable for developers. When your code is running across a dozen different types of processors, you need debuggers that can keep up. Who wins? Every engineer struggling to optimize or fix complex, accelerated applications. Watch for how quickly this LLDB architecture gets adopted by other hardware vendors. If it becomes a standard, it could significantly lower the barrier to writing software for this multi-chiplet, GPU-heavy future.

Finally, this brings us to a more philosophical arc, one about the culture of building itself. An essay this week tackled The Myth of Neutral Code. It addresses a recurring complaint in some developer circles: that software is being "radicalized," that politics is invading what was once a pure technical space. The essay's argument is that this view is itself a fantasy. Code is never neutral. From the algorithms that decide loan approvals or social media feeds to the very design of a programming language or a license, technical decisions encode values, priorities, and worldviews. The open source movement had its own politics. The choice of what to build, who to hire, and what problems to solve is inherently political.

This matters because it's the foundational cultural battle in tech right now. It's about whether the industry sees itself as a tribe of pure technicians or as a group of builders whose work shapes society, for better or worse. Who loses? The illusion that developers can be passive, that they bear no responsibility for the societal impact of their creations. Who wins? A more mature, reflective engineering culture that proactively considers ethics and equity, not as an add-on, but as a first principle. Watch for this tension to play out in conference topics, in hiring, and in the features that get prioritized. The tools we build are not just tools. They are the embodiment of our choices. And this week, from the mathematical scrutiny of law to the physics of the data center to the energy cost of a multilingual world, those choices and their consequences have never been clearer.

Welcome to the business and markets act, where this week the ledger of ambition shows some very large numbers, both in black and in red ink. We begin with a story of unraveling at the very top of the tech world. Nvidia, the company that became synonymous with the AI boom, closed the week below a four trillion dollar market cap for the first time in over a year. The immediate catalyst was the resignation of CEO Jensen Huang, but as the fallout continues, investors are grappling with more than just a leadership transition. The stock slide reflects a market reassessing the durability of a dominance that seemed, until very recently, unshakable.

And that reassessment gained new fuel from a separate, explosive allegation. A report emerged suggesting Nvidia didn't just benefit from the frantic demand for its AI chips, it may have actively engineered the scarcity. The claim, detailed in a new investigation, is that the company invested in AI startups specifically to fund their purchases of Nvidia GPUs, effectively creating its own demand and tightening the market further. If true, it paints a picture of a corporation not just riding a wave, but meticulously shaping the ocean itself. The combination of a sudden CEO departure and these serious market-manipulation allegations creates a profound crisis of confidence. It's a stark reminder that even the most meteoric valuations are built on a foundation of trust, and when that foundation cracks, the financial gravity can be swift and severe.

From the vertiginous heights of chipmaker dominance, we turn to the more terrestrial struggles of a startup aiming for the stars, or at least, the factory floor. Foundation, a humanoid robotics startup with notable backers, missed its payroll last month. Employees were still waiting for their wages as a CNBC report published this week detailed that the company was low on cash, even as its CEO had taken a multi-million dollar loan from the business. This is more than a simple liquidity crunch. It's a story about the gap between futuristic vision and present-day operational discipline. Building a general-purpose humanoid robot is arguably one of the most ambitious engineering challenges on the planet. It requires not just technical breakthroughs, but a mountain of capital and relentless execution. Missing payroll is often the first, unmistakable signal that the mountain is crumbling. It puts all other promises into question, and for a sector already scrutinized for its hype, it's a sobering case study in how quickly ambition can outpace reality.

But this week's economic ledger wasn't all about stumbles and scandals. We also got a powerful, retrospective look at an investment that paid off on a scale that's almost hard to comprehend. A new analysis found that COVID-19 vaccines generated up to eight hundred and nine dollars in economic benefits for every single dollar spent on their development and distribution. Let that number sink in. For every dollar invested, society saw a return of over eight hundred. The study tallies the value of lives saved, hospitalizations avoided, and lockdowns shortened, translating public health into a staggering economic multiplier. In a business context, it's the ultimate proof of concept. It demonstrates that strategic, mission-driven investment in foundational technology, in this case biomedical technology, can yield returns that dwarf even the most successful venture capital exit. It's a reminder that the most impactful ledger isn't always a corporate balance sheet, but the broader economic vitality that innovation can unlock.

And speaking of steady returns, one investment house that needs no reminder is Berkshire Hathaway, which this week broke out to a record high. The so-called Oracle of Omaha's conglomerate saw a fourth-quarter rally reach nineteen percent. In a week of tech volatility and startup drama, Berkshire's steady climb, built on insurance, railroads, and a timeless value philosophy, stands in quiet contrast. It doesn't make headlines for engineering scarcity or missing payroll. It simply compounds, a reminder that in business, there are multiple paths to creating value, and not all of them require reinventing the wheel or catching a rocket. Sometimes, the most radical act is simply to be durable.

So, what's the throughline this week in business? It's accountability. The market holding a giant accountable with a trillion-dollar markdown. The serious questions about how that giant played the game. A startup being held accountable by its own employees for the most basic promise of employment. And a retrospective analysis holding up a global scientific effort to the cold, hard light of economic accountability, finding it not just justified, but spectacularly so. The numbers tell the story, but the story is always about people, choices, and the sometimes immense cost of ambition.

Welcome back. If the first half of the show looked at what we build, this hour looks at what we discover. This week, discovery came with a rocket, a telescope, a brain scan, and a very old skull. Science is not one neat lane. It is a set of instruments pointed at very different problems, from the edge of the solar system to the edge of memory. Some findings are about what is out there. Some are about what is in us. All of them are about how close we are getting to seeing clearly.

Let's start with the most literal leap. SpaceX tried something that sounds like science fiction and behaved like engineering: catch both stages of Starship near the launch tower. The 15th test flight successfully reached orbit, which is already a major milestone. But the historic double catch fell short when the Super Heavy booster malfunctioned, forcing controllers to abort its return before it exploded. That failure is not a consolation prize. It is data. Catching a booster is not just a stunt. It is a way to make reuse faster, safer, and less dependent on long recovery operations. The near miss shows what ambition costs in real time. A successful orbit proves the vehicle can do its job. The aborted catch proves the next step is still hard.

While engineers were trying to catch one object in Earth's sky, astronomers were finding another way out past Pluto. A giant icy world temporarily named 2026 ED138 could become the sixth officially recognized dwarf planet. It was identified by an astronomer analyzing observations from the Simonyi Survey. This matters because the solar system is not a finished map. It is more like a neighborhood where the streetlights are dim. A new candidate this far out reminds us that classification is not pedantry. It is how we decide what counts as a planet, a dwarf planet, or a small body. You can follow that story through the candidate dwarf planet report.

At the same time, the European Space Agency released the largest dataset yet from its Euclid space telescope, with 1.7 petabytes of observations covering more than 900 million galaxies. That is not just a big number. It is a shift in scale. Modern astronomy increasingly works like a census. Instead of studying one galaxy at a time, scientists can compare hundreds of millions, looking for patterns in how the universe is arranged. The Euclid data release gives researchers a much larger map, and maps turn mystery into measurement.

Then we move from the cosmos to the brain. Stanford Medicine researchers report hidden brain signals that may predict Alzheimer's disease years before memory problems develop or harmful proteins begin accumulating in the brain. For families living with Alzheimer's, that is not just academic. It is a possible early warning. If risk can be spotted before symptoms, doctors may have a better chance to study the disease, enroll people in trials, and give patients time to plan. The promise is not a cure. The promise is a longer runway. The Alzheimer's prediction study points to that possibility.

And there is another brain related story that is more visual. Researchers from three institutions, including UC Santa Barbara, report an AI powered bionic eye breakthrough that could help blind people see again. The challenge is not just putting hardware in place. It is getting a brain implant to produce reliable visual experiences. AI appears to help with that translation problem. A bionic eye is not just a camera. It is a bridge between light and perception. The bionic eye research is a reminder that AI can do more than write emails.

From the brain, we go to the electron. Physicists captured electrons making seemingly impossible quantum jumps across energy gaps normally off limits, using a new type of slow motion electron camera. These jumps happen on timescales so short they usually blur together. Watching them is like learning how a machine works by seeing the gears move. The quantum jump experiment gives scientists a clearer view of processes that normally happen too fast to measure. That kind of view matters because quantum physics is not just theory. It is the foundation for future sensors, computers, and materials.

And then, because science can be both futuristic and ancient, forensic evidence from 5,000 year old skulls suggests some of ancient Egypt's earliest rulers had members of their royal households killed when they died, possibly so those individuals could continue serving them in the afterlife. That story is not cheerful, but it is important. It reminds us that the past was not a museum of pottery and pyramids. It was full of people making choices, some terrible, under systems of belief and power we are only beginning to understand. The ancient Egypt forensic study adds a human face, and a human cost, to one of history's most familiar civilizations.

So this is the science act in one line: we are learning to catch rockets, map galaxies, read brains, watch electrons, and listen to bones. Each sounds like a separate miracle. Put together, they show a pattern. Science is getting better at making the invisible visible. Sometimes that means a distant icy world. Sometimes it means a signal in the brain. Sometimes it means a jump inside an atom. And sometimes it means a 5,000 year old decision that still echoes in how we understand power, ritual, and death. The universe keeps giving us new things to notice. Our job is to build the tools, and the humility, to notice them well.

This week keeps returning to one question. What happens when our most elegant ideas meet the physical world? We build systems that feel weightless. We describe the cloud as if it were a sky, AI as if it were a mind, and markets as if they were just numbers on a screen. But the best stories remind us that abstraction always has a floor. It has walls, cables, energy bills, legal text, and human institutions underneath it. The theme is the tension between abstraction and reality.

Start with the law. The formalization of the U.S. tax code is not just a clever programmer’s project. It is an attempt to turn a messy human statute into something a machine can check, line by line, proof by proof. That is a powerful idea. It treats policy as code, and code as a place where bugs have consequences. But it also exposes a gap. Tax law is not only math. It is history, politics, interpretation, and compromise. The effort is valuable precisely because it shows where the abstraction can help and where reality will keep pushing back.

The same pattern appears in infrastructure. The story about the cloud as someone else’s computer cuts through the marketing language. The cloud is not a magical layer floating above the earth. It is a building with cooling pipes, power cables, and failure modes. When one data center overheats, the abstraction collapses. Applications, services, and routines depend on physical conditions. This matters because modern life has made the tangible invisible. We treat digital systems as if they cannot be touched, until they are.

Then comes the second theme, the cost of ambition. Nvidia’s market turbulence, including the fallout from its CEO resignation, shows how quickly confidence in a technology leader can reshape an entire sector. The company became a symbol of AI progress, and now its story is also one of governance, trust, and scarcity claims. Ambition is not free. It carries reputational risk, market risk, and the human problem of leadership.

SpaceX gives another version of the same lesson. The near miss on a double Starship catch was not a small engineering stumble. It was a reminder that the most futuristic ideas still have to survive physics, timing, hardware, and failure. Catching a rocket is not a demo. It is a negotiation with reality at enormous scale.

Across science and business, the thread holds. The Euclid data release, the energy demands of AI, the economic value of vaccines, and the financial strain at a robot startup all point to one truth. Progress is measured not only by what we imagine, but by what we can sustain. The week’s thread is not anti-technology. It is pro-reality. It asks us to respect the materials, institutions, and consequences beneath the abstraction.

And finally, for our one more thing, a delightful and wonderfully nerdy palate cleanser. This week, someone with a lot of patience and a deep love for classic adventure games put together a side-by-side comparison of all thirteen different versions of "The Secret of Monkey Island." That's right, from the original 1990 floppy disk release through countless ports and remasters, all the way to the latest editions. You can drag and drop any two versions onto the same screen and see exactly how the pixel art, the color palettes, and even the user interfaces evolved over decades. It's a beautiful, granular look at digital archaeology, showing how a single creative vision gets translated across changing hardware. It's the kind of project that reminds you that our digital culture has layers, like a very charming, pixelated onion. You can find that lovely comparison over at ReadFlow.

And speaking of layers, that brings us to the end of this week's show. From the abstract logic of tax code to the physical thrill of catching rockets, and from the hidden signals in our brains to the hidden ocean on a distant moon, it's been a journey. Our thanks to you for listening to ReadFlow Weekly. The producer is Anya Petrova. Our theme music is by Transistor Frequency. I'm your host, Alex Rivers. Until next week, take care.

And that brings us to the end of another ReadFlow Weekly. From the formal logic of tax codes to the fiery physics of a Starship landing, it’s been a week that reminded us our most ambitious systems are always grounded in a physical world, with all its costs and consequences.

Thank you for spending part of your Saturday with us. If any of these stories sparked your curiosity, you can dive deeper into the original reporting and analysis on ReadFlow. Save the articles that matter to you, and build your own understanding of the week that was.

We’ll be back next Saturday with another edition. Until then, keep reading. For ReadFlow Weekly, I’m your host. See you next week.