The Compiler Insight

The Correspondence

The file I’d written wasn’t documentation. It was a grammar. It defined what was valid and what wasn’t. Sections that contradicted the CANON couldn’t compile — they were structurally inconsistent, the way a program with a type error can’t compile. Sections that obeyed the CANON were guaranteed consistent.

Governance IS compilation.

Not metaphorically. Structurally. Define the constraints. Build the validators. Invalid content becomes impossible. Valid content compiles. The governance framework and the compiler are the same machine.

I’d spent 23 years writing academic papers, building clinical AI systems, running grants, supervising labs. An MSE in Systems Engineering at Penn. A PhD in Genomics under Junhyong Kim. An MD. A post-doc with Dwight Stambolian. Warren Ewens on my thesis committee. Then Stanford. Then UCSF. Then UCF. 65 peer-reviewed publications. Four clinical trials. $38M+ in funded research. And the insight came while writing a book about money and death on a Sunday afternoon in Orlando. The things you’re not looking for are the things that find you.

49 Commits in 5 Days

Once you see it, you can’t unsee it.

Monday: I restructured the entire manuscript around CANON.md files. Tuesday: I realized the pattern scaled. Wednesday: I started building validators. Thursday: I saw inheritance — child sections inherit parent constraints, the way child classes inherit methods. Friday: I had a framework.

49 commits between December 29th and January 3rd. Each one a step in the crystallization. The book was still being written, but the governance framework had taken on a life of its own — growing alongside the manuscript like a vine climbing a trellis, except the vine was becoming more interesting than the trellis.

CANNON

I called it CANNON. Two N’s. Because it felt like a weapon — something that could blow through the accumulated dysfunction of 23 years of ungoverned AI systems. Because the double N distinguished it from the governance file (CANON.md) it was built on. Because naming things at midnight on New Year’s Eve is not a discipline known for its precision.

The name would change. The insight wouldn’t.

Governance is compilation. Invalid states are compile errors. The specification IS the product. These weren’t decisions I made. They were facts I observed, sitting in a home office in Orlando, writing about precision medicine economics, discovering that the solution to every problem I’d spent my career failing to solve was hiding inside a manuscript about the problem itself.

The book that described the disease contained the cure.

Context

Type

Data

post

score-meter

score: 255, label: COMPILATION

A book about precision medicine economics accidentally invented a governance framework.

It was a Sunday. December 29th, 2025. I was writing a book.

Not a technical book — a book about dividends and deaths. About the economics of precision medicine. About what happens when the people who profit from healthcare and the people who die from healthcare occupy the same spreadsheet. The manuscript was called Dividends and Deaths, and it was growing faster than I could organize it.

Chapters bred sub-chapters. Arguments spawned counterarguments. Evidence piled up without structure, the way evidence does when you’re writing about something you’ve been angry about for a decade. The book was becoming ungovernable.

So I created a file called CANON.md.

The idea was simple: write down the rules. What does this chapter believe? What evidence supports it? What constraints must hold? Put the governance in a single file at the root of each section, and let the structure enforce itself.

I committed it. f58ad6d. 1:42 PM Eastern. The first CANON.md.

And then I noticed something.