Martian Moneyball
The scene everyone remembers from Moneyball is the scouting room. A dozen men who have watched baseball their entire lives are arguing about a player, and the arguments are about his swing, his confidence, his jaw, whether his girlfriend is a problem. Every one of those is a proxy. Not one of them is the thing they actually want to know, which is whether the man gets on base.
Billy Beane’s move was not statistics against instinct. It was noticing that an organization had been running on proxies for something it could measure directly, and that the proxies were cheaper to obtain and worse.
An org chart is a proxy too.
Headcount is a proxy for capability. A title is a proxy for authority. A policy document is a proxy for whether anyone is complying with anything. A resume is a proxy for whether someone can do the work. We use all of them for the same reason the scouts used the jaw: measuring the real thing directly was too expensive.
That is the part that changed.
Work as something you can run
Here is the claim, in one line: work scales as evaluation, not headcount.
Write the work down as math — terms, equations, substitution that preserves meaning — and then run it on a runtime made of three things:
- an agent, which does the work
- a record, which is append-only and cannot be quietly edited
- checks, which either pass or fail, and which anyone can run
Once that exists, the scarce input is no longer labor. It is the quality of what you wrote down. Specification becomes the cost function, and everything downstream of a good specification gets cheap in a way it never was before.
Why math and not a process document
A process document cannot be run. That is the whole objection.
You can write “the team will ensure invoices are reconciled monthly” and it will sit there being true or false depending on facts nobody is checking. The sentence has no interpreter. It is a promise about the world, dressed as a procedure.
The reason to write work as math is narrower and more boring than it sounds: you
want substitution to be legal. You want to be able to replace a thing with what
it means and have the meaning survive. There is no calculus in which x := x + 1
is an equation, and an organization built out of statements like that is an
organization you cannot reason about — only argue about.
Append-only records instead of mutable state. Locked claims instead of things that drift. Artifacts that give the same output for the same input, every time, for anyone who runs them. None of that is exotic. It is what makes a system something you can check rather than something you have to trust.
Compliance that compiles
The sharpest version of this is what happens to compliance.
Today compliance is an attestation. Somebody senior signs a page that says we certify that the following controls are in place, and the value of that page is entirely a function of how much you trust the signature. The regulator trusts the auditor, the auditor trusts the interview, the interview trusts the person, and at no point does anyone run anything.
The alternative is that an obligation goes in as a rule and comes out as a running system that emits its own proof. Not a document about the control — the control, executable, producing an artifact a third party can verify without asking you anything.
That is the difference between “we are compliant” and “here, run it yourself.” One of those is a claim. The other is a result.
The Martian problem
The other film worth stealing from is The Martian.
Mark Watney does the work of a space agency by himself. The interesting thing is that the book is not really about ingenuity. It is about method. He keeps a log, obsessively. He tests everything before he depends on it. He does not trust a result he has not verified, including — especially — his own.
He writes everything down not for posterity but because the record is how he catches his own errors. When the hydrazine blows up in his face, the log is what tells him why.
That is not a personality trait. It is an operating model, and it happens to be the same one: an agent, a record, and checks. He is doing institution-grade work without an institution, and the reason it holds is that he refuses to carry anything forward he has not proven.
Where I got this
I did not theorize this and then go looking for confirmation.
I built a production settlement platform on my own — no team — where the obligations went in as rules and came out as checks somebody else can run. Then I sat down and wrote out what had actually made that possible, because it clearly was not talent and it definitely was not headcount.
The theory came out of the practice. That order matters, and I would rather be judged on it than on the argument alone.
Where it breaks
Every theory should say where it strains, so here is mine.
The whole thing requires a conscious operator who decides what must remain true. The math will happily evaluate whatever you give it. It will not tell you which invariants matter, which claims are load-bearing, or when you are lying to yourself. Somebody has to stand outside the calculus and choose — and that somebody is an axiom, not a result.
Beane still had to decide that winning meant getting on base. The numbers did not hand him that. They only became powerful once he had.
Truth-seeking, in this frame, is not a virtue you bring to the work. It is a component. If the operator is not building machinery that hunts his own errors, the runtime faithfully evaluates a wrong specification and produces confident garbage at scale, which is worse than the org chart it replaced.
What it predicts
Four things, all near-term, all falsifiable, which is the only reason to bother writing them down:
- Solo operators delivering institution-grade work that used to require a department, at a quality that survives audit.
- Diligence moving from proxies to artifacts — fewer reference checks, more show me the thing running.
- Compliance becoming executable rather than attested.
- Org charts starting to look like legacy code: kept because rewriting them is frightening, not because anyone believes they are right.
The scope is deliberately narrow. This is about work made of bits, paper, money and law. It says nothing useful about manufacturing or anything else where physical throughput is the constraint. You cannot substitute your way out of a supply chain.
The formal version, with the actual math, is here: The First Scaling Theory.