The Pattern of the Swamp — And How to Drain It

I’ve now written about two systems that appear to have nothing to do with each other.

One is government IT — a 70% failure rate sustained for a decade, a 13% success rate on projects over $6 million, hundreds of billions of taxpayer dollars, and not one career ended by any of it. The other is family court — a venue where life-altering rulings are issued under a standard nobody can define, by actors nobody can sue, funded by incentives nobody audits.

Different buildings. Different personnel. Different statutes. Readers treat them as separate scandals.

They’re not separate. They’re the same organism wearing two different bodies. Same skeleton, swapped organs.

People have a name for this organism: the swamp. But years of promising to drain it should have taught us something — the swamp was never a place, and it was never a party. Places can be raided. Parties can be voted out. The swamp is a pattern, and a pattern just regrows in whatever host you move it to. You can’t drain what you haven’t diagrammed. So let’s diagram it.

The Five Vertebrae

Shielded actors. In government IT, it’s civil-service protection and pension math: the CIO who presides over a nine-figure failure gets reassigned, never fired. In family court, it’s quasi-judicial immunity: the officials whose rulings reshape childhoods are shielded from liability for those rulings. The organ differs; the function is identical. Failure has no cost for the people producing it.

Captured gatekeepers. In IT, the Big Integrators bid aggressive, change-order the contract to triple, deliver something contractually compliant that solves nothing — then win the follow-on contract to fix it, staffed through a revolving door with the agencies they bill. In family court, the family law bar performs the same role: the professionals who profit from conflict are the same professionals the system consults about its own rules. Insiders write the rules insiders are graded by.

Undefined standards. Government IT runs on phantom requirements — wish lists mistaken for specifications, so no delivery can ever be measured against intent. Family court runs on “the best interests of the child” — a standard with no operational definition, which converts every ruling into unappealable discretion. You cannot fail against a yardstick that does not exist. That is not a bug in either system. It is the load-bearing wall.

Money that rewards volume, not outcomes. More than $100 billion a year in federal IT spend, the majority of it propping up legacy systems whose continued existence justifies next year’s budget. In family court, Title IV-D incentive payments reimburse enforcement activity with no ceiling and no accuracy requirement — the state is paid for volume, not for getting it right. In both systems, the money flows toward the process continuing, never toward the problem ending.

Measurement resisted. When a tool arrives that could drop the cost of an honest look to nearly zero — and AI is exactly that tool — government IT responds with the resistance playbook: governance frameworks first, pilot committees, ethics reviews, delay dressed as diligence. Family court solved the problem earlier and more elegantly: it was simply never instrumented at all. No data, no baseline, no appeal to the numbers. In both systems, honesty ends careers, so honesty is the one deliverable that’s never funded.

Five vertebrae. Two bodies. Here is the whole anatomy on one page:

How to Read the Map

The middle column is the argument.

When a reader doubts the family court claims — and some will, because family court hides its numbers better than any agency in America — you don’t argue the right column. You point at the left one, where the receipts are federal: GAO has issued 1,881 IT recommendations since 2010 and 463 remain open. Then you ask one question: if the identical five mechanisms produce documented catastrophe in the domain we can measure, what do you suppose they produce in the domain designed never to be measured?

And notice what the congruence rules out. These two systems share no personnel, no statutes, no vendors. If the failure were about bad people, the pattern wouldn’t replicate across bodies that have no people in common. This is not a conspiracy. It doesn’t need to be. It’s what discretion without measurement always becomes, wherever it’s planted. The skeleton grows the same shape in any soil.

Half of the middle column already has a name in the economics literature. George Stigler won a Nobel for demonstrating, back in 1971, that regulation tends to be acquired by the industry it was built to constrain and operated for that industry’s benefit. That’s vertebra two — proven five decades ago for railroads and trucking, and visible today in every revolving door between the agencies and the integrators, every bar association consulted about the rules that feed it. What the map adds to Stigler is that capture never travels alone. It arrives with the shield, the fog of undefined standards, the volume-priced money, and the allergy to measurement — five bones of one skeleton. Economists have spent fifty years since Stigler debating whether capture can be regulated away. The map’s answer is that the question contains its own punchline: you’re asking the captured to perform the surgery.

The Output Is Permission

Follow the arrow at the bottom of the five rows. All five vertebrae discharge into a single result, and it’s worth naming precisely, because it’s the real product these systems manufacture.

Liberty without sovereignty is permission — revocable, conditional, granted by another power. You can feel free inside such a system the way a tenant feels free on someone else’s platform: free to post, sell, build — until the owner changes the terms. Then the freedom evaporates, because it was never yours. That was license, not liberty. The difference between a citizen and a subject, an owner and a renter, is not how free the good days feel. It’s who sets the defaults, and who can override whom.

The administrative state has quietly made tenants of us all. Your project succeeds if the integrator’s change orders allow it. Your family survives court if an unappealable reading of an undefined standard allows it. The founders ran the wiring the other direction — popular sovereignty first, rights that precede government and constrain it. Strip the sovereignty out and the rights don’t disappear; they degrade into something worse. They become wishes.

A republic where the people rent their liberty back from its administrators is not a republic. It’s a protectorate with elections.

X Versus DOGE

And we’ve been handed the cleanest natural experiment imaginable on exactly this point. Elon Musk drained one swamp and bounced off another, and the difference between the two is this entire article.

At X, he was the sovereign. He owned the record, the keys, and the defaults — so he could cut the workforce by roughly eighty percent, past every expert prediction of collapse, and the platform kept flying. The bureaucracy inside Twitter had no shield, no captured gatekeepers with independent power, no ability to rebaseline the record against its owner. Architecture let him.

At DOGE, he was a tenant. He walked into the skeleton with a mandate but no ownership — no append-only record, discretion on every side, every shielded actor and captured gatekeeper still holding their own keys. And by his own account, actually fixing the problem was the thing that would get him killed. I take him at his word on the fear and leave the forensics to him — but notice what the contrast proves either way. Same man. Same methods. Same appetite for the fight. The only variable that changed was which column of the map he was standing in.

That’s not a story about Elon. That’s a story about the map. The swamp doesn’t defeat reformers because they’re weak; it defeats them because they enter as tenants. Sovereignty was the whole difference between a turnaround and a threat assessment.

The Inversion

Now the bottom of the map — and count carefully. Five vertebrae going in. Four inversions coming out.

That asymmetry is not sloppiness; it’s the whole design insight. The first and fifth vertebrae — shielded actors and resisted measurement — collapse into a single fix, because a shield only works in the dark. Immunity, tenure, and pension math survive contact with the public exactly as long as the record can be rebaselined, renamed, and rewritten. Make the record append-only and the shield doesn’t have to be outlawed. It simply stops working.

That is the difference in kind between the SQUEIL Constitution and every reform proposal you’ve read. Reform asks the administrators to behave better. Architecture removes the behavior from their discretion:

The append-only record means history cannot be rebaselined — every decision, every change order, every ruling of the platform’s own governance sits in a chain that its authors cannot edit. Cryptographic tenant sovereignty means capture has nothing to seize — no administrator holds the keys to what tenants own, so there is no gatekeeper worth revolving a door into. Published objectives mean the standard exists before the work does — recovery targets and specifications on the record in advance, so “compliant but useless” has nowhere to hide. And un-amendable stakeholder rights mean the incentives are locked beyond the reach of the people they constrain — the one clause every captured system is missing, because every captured system lets the insiders amend the rules that bind insiders.

Madison understood this in 1787: if men were angels, no government would be necessary — and since they aren’t, you build auxiliary precautions, structure that doesn’t depend on virtue. The administrative state abandoned that insight somewhere along the way. It runs on discretion and asks for trust. The SQUEIL Constitution is a Madisonian move executed in code instead of parchment.

Same Autopsy, Rival Treatment

Before I tell you what I want, I owe you the strongest objection to everything above — and it comes from the person who has stared at this pattern longer and closer than almost anyone.

Jennifer Pahlka founded Code for America, served as deputy CTO of the United States, and in 2023 published Recoding America, the best book yet written on why government fails at delivery. Her autopsy matches this map nearly row for row: integrators capturing procurement, requirements cascading down from statute without ever meeting a user, implementation treated as beneath the dignity of policy, agencies that outsourced their own competence until nobody inside could evaluate what they were buying. Put her book next to my government IT essay and you’ll find a progressive civic technologist and a seasoned government IT specialist, starting from opposite politics, describing the same corpse. That convergence is itself evidence. When ideological opponents produce the same anatomy, you’re looking at bone, not spin.

But her prescription runs opposite to mine. Pahlka’s answer is to rebuild capacity inside — hire real technologists into government, give implementers a voice in policymaking, repair the pipeline until the state can deliver again. Mine is to build the alternative outside, on architecture the state can’t bend.

Here’s my honest answer to her, and it isn’t a dunk. I hope she wins. A competent state is better than an incompetent one, and I intend to sell to it either way. But capacity poured into unchanged structure becomes a better-staffed version of the same discretion. The shield and the dark don’t care how talented the engineers are; the five vertebrae have digested waves of gifted reformers before, and the map explains why — talent changes what a system can do, architecture changes what it must do.

The two paths aren’t even exclusive. Internal reformers negotiate hardest when a credible alternative exists outside the walls; monopolies reform under competition or not at all. And there’s a clean experiment that separates our theories. Staff the agencies with Pahlka’s people, then offer them append-only records and published objectives. If missing capacity was the disease, they’ll adopt the measurement eagerly. If the skeleton is the disease, they’ll resist it no matter who’s hired. Watch which one happens.

What I Actually Want From Government

Here’s where I’ll surprise the readers who think they have me pegged.

I don’t want the government abolished. I’m building to sell to the federal government — I don’t hold a contract yet. I want it demoted — from landlord to counterparty.

Out of the way as a sovereign over citizens: no permission required to build parallel systems, no certification regimes that quietly become moats for the incumbents who broke everything, no securities frameworks from 1933 strangling instruments of sovereignty they were never written to comprehend. But still in the room — as a customer, and as a referee. A government that buys from sovereign systems, and enforces contracts between them, is doing the two jobs a free republic actually assigned it.

And frankly, a government that wanted to fix itself could do worse than to study the bottom block of this map. Every one of the four inversions would work inside an agency tomorrow. That none of them will be adopted voluntarily tells you which column the agencies know they belong in.

The Test Runs Both Ways

Two invitations before the close.

First: run the map yourself. Pick any institution — zoning boards, occupational licensing, university accreditation, hospital billing. Fill in a third column. If all five rows populate, you’ve found the same animal, and I want to see the column you built.

Second — and this is the row I refuse to skip — run the map on SQUEIL. Every extractive platform in history launched with a liberation pitch, and success bent every one of them. The pattern does not care whose logo is on the door, including mine. Which is exactly why the Constitution’s constraints bind the founder first: the append-only record applies to my decisions, and the rights it guarantees are locked beyond my reach. The sovereign who recognizes a higher law constrains himself — that is precisely what makes his tenants free. The day SQUEIL fills a third column on this map, the map convicts us too. I built it that way on purpose.

The Bottom Line

Two systems, no shared parts, one skeleton. That congruence is the strongest evidence I can offer that America’s dysfunction is not a personnel problem, not a party problem, and not fixable by asking the pattern to please stop. Patterns don’t respond to asking. They respond to architecture. That is what draining the swamp actually requires — not new occupants for the old structure, but structure the occupants can’t bend. Drain the discretion, and the water goes with it.

A free republic is not a country with no government. It is a country where citizens own their liberty instead of renting it — where the defaults are ours, and the burden of override falls on power instead of on us.

---

The current system cannot fix itself — the map above is the anatomy of why. The SQUEIL Constitution is the inversion built and running: append-only records, cryptographic tenant sovereignty, published recovery objectives, and stakeholder rights that cannot be amended away by the people they constrain. Read the pattern. Then read the alternative. Then decide which column you want to live in.