A qualification exam disguised as a book

Around 1995, we hid the messy details behind abstractions. This freed software development to advance rapidly. But it also hid the constraints that shape high-performance systems. The tradecraft of reasoning about systems as a whole, the skill that built the world's fastest supercomputers, became tacit knowledge passed from person to person and never written down.

That knowledge is now retiring out of the workforce.

This book exists to transmit the method of transmission. If you want to teach this material, you need to know how it was taught. If you want to learn it, you need to know what you are committing to.

What This Book Does

The book performs what it teaches. Each chapter demonstrates a technique while explaining it. By the end, you will have experienced the formation process that the larger books in this series employ throughout.

You will learn to model your own thinking and articulate it to others. You will learn to spot transcendent patterns, solutions that appear independently across different contexts because similar constraints shape similar solutions. You will see how the attention mechanism in modern AI maps to the same cognitive processes you already use.

None of this requires mathematics. The concepts are separated from their implementations.

What This Book Is Not

This is not an introduction to high-performance computing in the conventional sense. There is no parallel programming tutorial, no cluster architecture overview, no benchmark analysis.

This is the cognitive foundation. The skills that made Cray Research engineers able to accomplish what nobody else on the planet could match. The orientation that treats "it has never been done before" as a starting point rather than a barrier.

Who This Book Is For

If you skim, you will get nothing. The material is designed to be experienced, not surveyed. That is not a sales pitch. It is a structural fact about how the teaching works.

The book serves as qualification for the two large books that follow: Nobody but Us (how Cray Research built the world's fastest supercomputers, from the software-engineering point of view) and The Wizard's Lens (how to observe and characterize AI systems using the same tradecraft). Those books require significant commitment. This one lets you make an informed choice.

If you are a professor considering whether to teach this material, the book shows you exactly how the pedagogy works. If you are a practitioner wondering whether Cold War-era engineering tradition has anything to offer modern AI work, the book demonstrates the connection directly.

The Appendices

Two declassified NSA documents from 1952-1953 show the distinction between labor-saving machines and revolutionizers: computing systems that enable work previously impossible to consider. This is not historical decoration. It is the operational definition of high-performance computing that this series transmits.

The final appendix is the road map for all six books in the two series, showing where this book fits and what lies beyond.

The Commitment

Approximately 20,000 words. Readable in about three hours if you engage fully. The free sample includes the opening chapter and the road map appendix, enough to know whether you belong here.

When second place meant people died, the answer was Cray Research.

When laboratories competed for the few scientists with Top Secret clearances, the best "bragging rights" brought in the best scientists. The ultimate bragging right was Seymour Cray's first and only CRAY-1 supercomputer. Each lab shot down the other's funding. In a boss move, Cray gave away the computer. The second customer walked in and paid cash. Accomplishing the apparently-impossible became the Cray Research mystique.

Supercomputing Comes From Codebreaking

Early radio technology created an invisible battlefield where codebreaking meant lives saved. Military cryptanalysis became the direct path to supercomputing. With lives at stake, existential pressure created cognitive frameworks that made the impossible routine.

How We Dealt With Overwhelming Complexity

A Cray Research veteran narrates how pioneers handled overwhelming complexity: pattern recognition anticipating Midway, systems thinking inventing magnetic core memory. Experience from one domain, applied in a new way, shaped supercomputing.

Through historical narrative and technical examples (debugging unreproducible problems, optimizing tightly-constrained hardware, bare-metal skills becoming lost to time), you can develop the same abilities. The author treats mastery not as a linear path, but as cycles building on previous cycles of mastery.

Today's Overwhelming Complexity

The Cold War stakes are gone but overwhelming complexity intensifies. The generation that developed these frameworks and devices is retiring and passing away. This knowledge has never been systematically transmitted to the next generation until now.

The path to Cray Research-level wizardry exists. It has been traveled. You can travel it today.

What if you could see how AI actually thinks?

Not metaphorically. Not theoretically. But through a working demonstration you can build yourself using physical objects—the same approach Donald Michie used in 1961 to prove machines could learn, by teaching matchboxes to win at tic-tac-toe.

The Wizard’s Lens reveals something that does not exist elsewhere in AI literature: a complete Large Language Model you can construct and operate with terrain maps, tokens, and attention mechanisms made tangible. This is not a metaphor for understanding AI. This is AI, demonstrated through physical implementation.

While others struggle to use AI effectively, this book teaches you to think like AI—to see the patterns, understand the mechanisms, and apply insights that enable you to accomplish what has never been done before.

The Hidden Knowledge

What Cray Research accomplished is well documented: we built the world’s fastest supercomputers and changed what was computationally possible. How we did it has never been written down—until now.

This book shares the systems thinking approaches and revolutionary mindset from that era, applied to modern AI. Not as history, but as practical methods you can use today. The same approaches that created computational breakthroughs in environments where second place was not survivable now reveal how to work with AI in ways others cannot replicate.

For those who recognize the significance of Cray Research: yes, this is that knowledge. For everyone else: you’re learning approaches that have already proven they enable the impossible.

What You’ll Discover

Through hands-on demonstrations and clear explanations, you’ll learn:

Token Context and Embeddings - Build a working model that shows how AI represents and processes information, using physical tokens and terrain maps that make abstract concepts concrete.

Attention Mechanisms - Understand how AI focuses on relevant information through a demonstration you can manipulate yourself, revealing why certain approaches work and others fail.

The Ping Pong Effect - Move beyond one-shot prompting to develop collaborative relationships with AI that grow more effective with each interaction.

Pattern Recognition at Scale - Learn to see how AI connects disparate information, and how to structure your work to leverage these connections.

Revolutionary Thinking - Develop the mindset that enables you to take on challenges others consider impossible, treating “it can’t be done” as an invitation rather than a boundary.

Why This Book is Different

Most AI books teach you to write better prompts or explain transformer architecture with equations. This book shows you how the mechanisms actually work through physical demonstration, then teaches you to apply that understanding in ways no one else can.

You’ll learn to use AI like nobody before you—not through tricks or techniques, but through genuine understanding of how Large Language Models process information, maintain context, and generate responses. This understanding transforms how you collaborate with AI, opening possibilities that others believe require technology that doesn’t exist yet.

Who This Book is For

This book serves multiple audiences:

Professionals and knowledge workers who need to accomplish complex tasks with AI and recognize that basic prompting has fundamental limitations.

Developers and technical practitioners who want to understand LLM internals without drowning in mathematics, gaining practical insight that informs better implementation decisions.

Strategic thinkers and innovators who need to solve problems that have never been solved before, and recognize that revolutionary results require revolutionary approaches.

Anyone who suspects there’s more to AI than what current tutorials and guides reveal, and wants to develop capabilities that create genuine competitive advantage.

The Author’s Background

Edward W. Barnard spent years at Cray Research during the era when we were accomplishing what had never been done before in computing. This book shares the systems thinking approaches and revolutionary mindset from that time, now applied to AI collaboration.

These aren’t theoretical frameworks. These are battle-tested methods from environments where results mattered more than credentials, where impossible challenges were solved through clear thinking and revolutionary approaches to complex systems.

The cryptographic origins of modern computing—from Alan Turing through Seymour Cray—created specific ways of thinking about information, context, and computational possibility. This book teaches those approaches in a form you can apply immediately.

What’s In It For You

Immediate capability - Learn to accomplish tasks with AI that others consider beyond current technology, through understanding how LLMs actually process and generate information.

Deep understanding - Build a working LLM yourself, making abstract concepts concrete and revealing why certain approaches succeed while others fail.

Transferable skills - Develop systems thinking approaches that work across all AI platforms and remain valuable regardless of how technology evolves.

Revolutionary mindset - Learn to approach impossible challenges the way they were approached at Cray Research: as interesting problems to solve rather than boundaries to accept.

Long-term mastery - Create a foundation for continually improving your AI collaboration skills, based on understanding rather than memorized techniques.

Whether you’re drowning in operational complexity, struggling with projects that seem beyond AI’s current capabilities, or simply know there’s a better way to work with these tools, this book will transform not just what you can accomplish with AI, but how you think about what’s possible.

The revolution isn’t coming. It’s here. The question is whether you’ll be among those who can see it.

Uncomfortable Truth

We send our young computer scientists and software engineers out into the world thinking the ACM Code of Ethics protects them. It does not.

This book confronts an uncomfortable truth: the corporate world our graduates enter has deliberately stripped itself of ethical guardrails. Through a combination of personal experience and careful historical analysis, we trace how this happened, and what it means for you.

What You Will Learn

• How Milton Friedman and Harvard Business School's Michael Jensen dismantled the mid-century view that CEOs should have an enlightened social conscience, replacing it with amoral shareholder value maximization
• Why the power imbalance between early-career employees and Big Tech titans is not an accident, and why it disproportionately affects women and persons of color
• What the military's Return With Honor framework teaches us about surviving situations where the rules are stacked against us, without becoming corrupted ourselves

The Author's Unusual Qualification

The author completed full SERE (Survival, Evasion, Resistance, Escape) training at the Air Force Academy in the summer of 1976, with training freshly updated based on the just-closed Vietnam Prisoner of War experience. That training addressed surviving illegal treatment under the Geneva Conventions through a framework of personal integrity and mutual support.

Decades later, working in high-performance computing, the author recognized something familiar: employees facing an amoral power structure with no enforceable ethical standards to protect them.

A Work in Progress

This book currently presents two chapters as urgently important background material. The solution is not to fix the power imbalance. The solution is to develop skills, on a personal level, to survive that imbalance and remain intact. That work continues.

The Two HPC Tradecraft Series

What is "HPC tradecraft" as I know it? This is high-performance computing (HPC) arising from signals intelligence (code breaking) needs. My particular lineage begins with the "Tunny" code breaking project at Bletchley Park during World War II. That produced such an odd-looking contraption that the "Wrens" operating it named it the Heath Robinson. W. Heath Robinson was a cartoonist known for portraying fantastically complex machines designed to accomplish simple tasks.

Thus the high-performance tradecraft I learned within Cray Research already had a 40-year history. Perhaps because so much was classified Top Secret, we never wrote down how we did things. Institutional knowledge was passed from person to person as mentorship, collaboration, and informal apprenticeship.

I created two book series to teach you HPC tradecraft.

If you are interested in learning HPC tradecraft, I invite you to begin with the first book HPC Tradecraft for Computer Scientists: What We Stopped Teaching.

If you are here for ethics when the stakes are unexpectedly high (either as teacher/mentor or student/early-career), start here with High-Stakes Ethics and continue through Unexpected Histories.

1. HPC Tradecraft Apprenticeship, three books, is the institutional tradecraft as I learned and practiced, and still practice. Tradecraft is useless unless replicable. The first book teaches my replication method; the second teaches how we accomplished what we did; the third teaches systems thinking, with close observation and characterization of HPC systems under load, using modern AI as the example. The three books are transmission protocol; HPC design; HPC operations.
2. The HPC Tradecraft Master Practitioner, three books, is the author letting loose and having fun, expressing his mastery of HPC tradecraft. High-Stakes Ethics prepares for the insurmountable power imbalance of a new graduate in the amoral corporate environment. Unexpected Histories demonstrates debugging skills applied outside computer science, showing their value and timeless nature. Constraint-Based Design demonstrates 1986 tradecraft directly applies to modern AI transformer design and usage.

What happens when institutional power is challenged?

What happens? People die.

This book applies the skills that debug code to debug history. Pattern recognition and root-cause analysis that define high-performance computing tradecraft are skills that apply equally to historical puzzles most people accept at face value. The results are unexpected, occasionally unsettling, and consistently illuminating.

The facts have been hiding in plain sight.

The Day the Internet Died

How do large systems innovate, and how do they fail? Big Tech presents itself as unprecedented and unstoppable. History suggests otherwise.

Three chapters examine moments when technology and infrastructure systems reshaped society, drew institutional response, or collapsed under their own scale. The 1988 Morris Worm revealed how fragile the early internet really was. The author watched the response unfold in real time. The coming antitrust reckoning for Big Tech replays earlier infrastructure battles, with the Department of Justice citing a 1958 railroad case as precedent. And the transcontinental railroad itself demonstrates that antitrust enforcement follows infrastructure, not innovation.

The First Silicon Valley Unicorn Startup

The Department of Justice cited one case when announcing its Big Tech investigation: Northern Pacific Railway v. United States (1958). Why would a famous railroad matter to Silicon Valley?

This narrative gleefully traces the legal and historical threads connecting railroad tycoons to tech titans. Along the way, we encounter the first Silicon Valley "unicorn" (a Mormon merchant in 1848 San Francisco), a famous lawyer who trolled the California Bar Association by writing a novel about trolling them, and the uncomfortable fact that Google and Facebook have hired the Pinkertons, the same agency that killed striking miners at Blair Mountain.

The Secret in Arlington

For nearly fifty years, a secret message lay hidden on a tombstone in Arlington National Cemetery. The grave belonged to William Friedman, founding cryptographer of the NSA, and his wife Elizebeth, who broke Nazi Enigma codes during World War II. In 2017, a cryptographic historian noticed something odd. Some letters had serifs, others did not. The widow had encoded her husband's initials using the same Bacon cipher they'd studied together in 1918.

Three essays explore hidden messages, buried truths, and the challenge of separating fact from carefully crafted fiction: from Elizebeth Friedman's classified wartime work (which J. Edgar Hoover kept secret because he'd claimed credit himself) to Wikipedia's subtle distortion of the military Code of Conduct, to the author's personal ties to accused Salem witches and their defenders.

Safe Havens Then and Now

For thirty-five years, St. Paul, Minnesota operated as an open sanctuary for criminals. Police Chief John O'Connor's "Layover Agreement" welcomed John Dillinger, Ma Barker, and Babyface Nelson, so long as they committed no crimes within city limits.

Three essays explore protected spaces and hidden origins: the Age of Sail terminology we still use without understanding, the parallel between gangster-era protection rackets and modern tech platforms (Google's reCAPTCHA forces users to train AI; Twitter's bot-reporting mechanism may train the platform to deploy bots rather than eliminate them), and why the origin of digital computers remained classified for decades.

Top Secret Ultra

The CRAY-1 supercomputer didn't emerge from nowhere. Its origins trace back to a classified team of World War II code breakers who quietly relocated to a glider factory in St. Paul after the war ended. They invented the first stored-program computer, but the NSA kept it secret for decades.

This section reveals the hidden history connecting wartime cryptography to the supercomputing revolution, from Engineering Research Associates' drafty factory to Seymour Cray's remote laboratory in Chippewa Falls, from the ENIAC patent disputes to the CDC 6600, the world's first supercomputer, built by just 34 engineers. IBM's Thomas Watson Jr. famously demanded to know how his vast organization had been bested by such a small team. Cray's response became legend: "It seems like Mr. Watson has answered his own question."

The Two HPC Tradecraft Series

What is "HPC tradecraft" as I know it? This is high-performance computing (HPC) arising from signals intelligence (code breaking) needs. My particular lineage begins with the "Tunny" code breaking project at Bletchley Park during World War II. That produced such an odd-looking contraption that the "Wrens" operating it named it the Heath Robinson. W. Heath Robinson was a cartoonist known for portraying fantastically complex machines designed to accomplish simple tasks.

Thus the high-performance tradecraft I learned within Cray Research already had a 40-year history. Perhaps because so much was classified Top Secret, we never wrote down *how* we did things. Institutional knowledge was passed from person to person as mentorship, collaboration, and informal apprenticeship.

I created two book series to teach you HPC tradecraft.

If you are interested in learning HPC tradecraft, I invite you to begin with the first book HPC Tradecraft for Computer Scientists: What We Stopped Teaching.

If you enjoy "once you see it, you cannot un-see it" stories based on researching primary sources, you are in the right place. This book is tradecraft practice spanning several centuries, demonstrating the timeless skills independent of any particular technology.

1. HPC Tradecraft Apprenticeship, three books, is the institutional tradecraft as I learned and practiced, and still practice. Tradecraft is useless unless replicable. The first book teaches my replication method; the second teaches how we accomplished what we did; the third teaches systems thinking, with close observation and characterization of HPC systems under load, using modern AI as the example. The three books are transmission protocol; HPC design; HPC operations.
2. The HPC Tradecraft Master Practitioner, three books, is the author letting loose and having fun, expressing his mastery of HPC tradecraft. High-Stakes Ethics prepares for the insurmountable power imbalance of a new graduate in the amoral corporate environment. Unexpected Histories demonstrates debugging skills applied outside computer science, showing their value and timeless nature. Constraint-Based Design demonstrates 1986 tradecraft directly applies to modern AI transformer design and usage.

This is Book 3 of The HPC Tradecraft Master Practitioner series, a capstone demonstrating that pre-1995 bare metal design methods remain directly relevant to modern AI architecture.

In 1986, the author wrote a text adventure game as a bare metal device driver inside the Cray I/O Subsystem. The game, Swiss Adventure, was inspired by Colossal Cave Adventure and written during a Cray Research training class. The source code architecture turns out to be isomorphic to the 2017 "attention is all you need" transformer architecture. Similar constraints shape similar solutions.

This book is a work in progress. The current release covers the user experience: what the game does, why it was designed that way, and what the adventurer learns by playing it. The game itself is live and playable at ewbarnard.com with no registration or data collection. The 1986 assembly language source and its PHP port are available on GitHub.

This book is not a normal code walkthrough. This book models the expert thinking common to HPC tradecraft as practiced in the Cray Research era. Typical IT books with sample code work through from one topic to the next. This book builds out a system layer by layer, first designing the user experience, then designing a way to capture and encode the user experience, and then taking advantage of bare metal features and constraints. The companion website allows you to execute the code (by playing the game) while observing the event trace and AI pattern trace moving from request to response.

This book requires a strong working knowledge of Part II of the earlier book Nobody but Us: A History of Cray Research's Software and the Building of the World's Fastest Supercomputer. Part II introduced the necessary concepts for working on bare metal.

• The live game: https://ewbarnard.com/
• The source code: https://github.com/ewbarnard/SwissAdventure