Corruption Is a Rotation

Contents

1. Reality is never corrupt at the beginning
2. The only door in
3. Two corruptions, and we only legislated one
4. Why nobody inside can feel it
5. A walk-down, applied to a charter
6. The case: an instrument built to protect an idea
7. Boeing, briefly and carefully
8. Why the repair everyone reaches for cannot work
9. What to watch, including at Blender
10. The claim, and where it breaks

Last time we read a structure by what it made impossible, and found a charter holding a denominator perpendicular to the thing everyone else was measuring. This lesson is the same structure running the other way. Charters fail. Constitutions get interfered with. It happens constantly, in every institution any of you will ever join, and the striking thing about it is how little of it involves anybody doing anything wrong.

I want to give you the mechanism precisely, because corruption is the subject on which otherwise careful people become lazy. The lazy account is that bad actors arrive and take things. That does happen. It is rare, it is illegal, and it is not what takes institutions apart. The thing that takes institutions apart is something for which we have no crime, no clause, and no alarm — and in our notation it has an exact shape.

Corruption is a rotation.

01Reality is never corrupt at the beginning

Start where the equation starts. An Actualizer — and remember the entity need not be a person; it can be a foundation, a firm, a federation, a government — receives a Reality. Actual over Expectation. The numerator arrives out of the Immutable Past, finished. The denominator is brought by the comparison machine before anyone is asked.

Neither of those is corruptible. This is not a moral claim, it is a structural one. There is nothing in A to bribe and nothing in E to lean on. At the instant it arrives, a Reality is exactly what it is, for everyone, always. The freshly chartered organisation and the hollowed-out one are in identical position on the day each of them opens the mail.

So if nothing on the right-hand side can be got at, and corruption plainly exists, there is only one place it can be entering from.

02The only door in

The one freedom an Actualizer has is on the left. Reality arrives, and more than one action remains possible. That is the whole of the liberty in this system and it is not small — it is the only thing anyone has ever had.

An entity can act however it wants. It cannot reach into either term, not with money, not with authority, not with a very good lawyer. What it can do is behave, and behaviour leaves a record.

Consciousness does not manufacture the future. It manufactures history — a collection of artifacts — and artifacts become Actual, and only Actuals retrain prediction. So the denominator does move. It is always moving, chasing a numerator that will not hold still. It simply never moves because somebody moved it.

Now watch what that implies. If an entity acts a certain way once, the record barely notices. If it acts that way for eleven years, the record is made of that behaviour. And the prediction machine — every prediction machine, inside the institution and outside it — has been trained on that record, faithfully, doing exactly its job.

At which point the behaviour is no longer a departure from what is expected. The behaviour is what is expected. P has learned it. Meanwhile the founding idea — the charter, the manifesto, the thing framed in the lobby — has been fed nothing at all, because it was never an Actual, and only Actuals retrain. It is still in the denominator. Its share of the denominator has just been collapsing for a decade.

That is the rotation. θ walks off the Ideal and down toward the real axis, and at no point in the walk did anyone touch the denominator, because that is impossible.

03Two corruptions, and we only legislated one

Which gives us two entirely different failures that share a word, and the difference is worth more to you than anything else in this lesson.

Figure 01 The one we prosecute, and the one that actually does the damage

	Fraud	Drift
What is attempted	Fraud · Overwriting the numerator. Putting a desired outcome where the record goes.	Drift · Nothing is attempted. Ordinary actions, each defensible on the day.
Does it work	Fraud · No. The Immutable Past is not editable — only the story about it, and stories conflict with artifacts.	Drift · Completely. It is the system operating correctly.
Intent required	Fraud · Yes. Someone has to decide to do it.	Drift · None. It proceeds faster when everyone is sincere.
Detectable	Fraud · Yes, eventually — two artifacts disagree and the disagreement is the evidence.	Drift · Not from inside. Every artifact agrees with every other artifact.
Felt by the entity	Fraud · Yes — it lands in the magnitude, as fear and then as a jolt.	Drift · No. It lands in the angle, and the angle is unfelt.
Remedy we built	Fraud · Statutes, auditors, prosecutors, an entire profession.	Drift · None. There is no act to prohibit.

Note what the first column really is. Overwriting the numerator with a desired outcome is not corruption’s clever version — it is the engine of fear and hope, running inside an organisation instead of inside a person. And it fails for the reason it always fails. You cannot edit the Immutable Past. You can only produce a second artifact that contradicts the first, and now there are two things in the record and one of them will be found.

The second column breaks nothing. That is precisely why it works.

04Why nobody inside can feel it

Here is the part that ought to disturb you, and the reason I keep dragging this class back to the complex log.

Corruption in the second sense does not change |E| much and need not change A at all. It changes θ. And the angle is in the part of surprise no instrument in the body reads.

The angle does not change how much you were paid. It changes which way the payment turns you. And you cannot feel a turn.

So when you ask the people inside a drifting institution whether anything has changed, they say no — and they are telling you the truth. Nothing they can feel has changed. The magnitude is unremarkable. Every quarter lands about where the last one landed. The jolt is absent because there is no jolt to have.

They are not lying to you and they are not lying to themselves. They are reporting, accurately, on the only component available to them. A rotation is by construction invisible from inside the thing being rotated, which is why the honest people are so often the last to know, and why the outside observer who has not been turned finds the whole thing obvious and cannot understand why nobody will admit it.

You met this exact structure last week wearing different clothes. The feudal lord could not feel the merchant. Not from inattention — from geometry.

05A walk-down, applied to a charter

And now the mechanism of the walk itself, which I have already given you in another context and will simply point at here.

In The Conservation of Surprise the identity is that intermediate expectations cancel. Only the first and the last survive:

Investor relations uses this deliberately over ninety days. An institution does it to its own charter over thirty years, and nobody is running it.

Each step is small. Each step is defensible on the day it is taken, by reasonable people, for good reasons, usually in a crisis, usually as an exception. No single installment is a scandal, because each one is smaller than perception. The line was never crossed. The line was moved, in increments beneath the resolution of anyone standing next to it.

And then one day somebody lines up the first expectation against the last and the distance is enormous — and there is no moment to point at, no meeting where it was decided, no villain to name. Only the schedule. The total was always going to be the total.

06The case: an instrument built to protect an idea

I want a case where nobody did anything wrong, because a case with a criminal in it teaches the wrong lesson. So: Mozilla. And I am going to be careful, because the temptation to moralise here is strong and moralising would obscure the mechanism.

The Mozilla Foundation is a California nonprofit public benefit corporation, a 501(c)(3) public charity, launched in July 2003. Its manifesto, written in 2007, says the internet is a global public resource that must remain open and accessible, that individual security and privacy are fundamental and must not be treated as optional, and that a balance between commercial profit and public benefit is critical. Read those again in a moment. They are still the text.

In August 2005 the Foundation created the Mozilla Corporation — a wholly owned, taxable subsidiary. And here is the sentence I want you to sit with, because it is Mozilla’s own, from the announcement, and it is not cynical in the slightest:

Wholly owned, to ensure that the Mozilla Foundation’s principles will be the guiding force behind the Mozilla Corporation.

That is an instrument built to protect the idea. It was designed by people who had thought about exactly this risk and had put in place exactly the structure they believed would hold. Hold that thought against everything that follows, because nothing that follows is a betrayal of it.

Figure 02 The angle walking, 2003 to 2026

Now the record, and only the record.

Figure 03 Concentration, and reach

Royalty revenue, FY202473.2%

Everything else26.8%

Firefox desktop share, 200932.2%

Firefox desktop share, 20265.3%

• Share of total revenue · audited
• Browser share · StatCounter

The audited statements carry a note headed Concentrations of Risk and Significant Customers. It reads: approximately 86% and 85% of Mozilla’s revenues from customers with contracts were derived from one customer for the years ended 2024 and 2023. The statements never name that customer anywhere. Mozilla’s chief financial officer, testifying in the Google antitrust remedies proceeding, put the dependence plainly — that reduced payments could start a downward spiral of usage as people defected, which could at the end of the day put Firefox out of business.

I am quoting their own disclosure and their own testimony on purpose. It is fairer than any characterisation I could write, and it establishes the thing that matters: they know. This is not a story about people who cannot see. It is a story about an entity whose available actions narrowed until the ones that remained all pointed the same way.

Figure 04 Artifacts, in order

1. July 2003 · August 2005

   The charity, then the instrument to protect it

   Foundation launched as a 501(c)(3). Two years later a wholly owned taxable subsidiary is created so that revenue — described at the time as “an unintended but real by-product” — can be handled without compromising the parent, and wholly owned so the Foundation’s principles remain the guiding force.

2. 2007, amended 2013 and 2018

   The manifesto is written and extended

   Ten principles. Privacy is added to Principle 4 in a later revision; “commercial goals” becomes “commercial profit” in Principle 9; four further commitments are published as an addendum in March 2018. The document grows. Nothing in it is ever relaxed.

3. 2009 onward

   Search royalties become the revenue

   Default-search placement is renewed, renewed again, and becomes the overwhelming majority of income. Each renewal is a sensible act by competent people. There is no meeting at which anyone decides to become dependent.

4. June and July 2024

   An advertising company, and an advertising feature

   Mozilla acquires Anonym, founded by two former Meta advertising executives, describing it as a trailblazer in privacy-preserving digital advertising. Weeks later Firefox 128 ships Privacy-Preserving Attribution, enabled by default with an opt-out. A privacy group files a complaint in Austria; no regulator has ruled on it. Mozilla later removes the feature and states it was never activated beyond its own trial.

5. October 2024 · February 2025

   The advocacy arm is cut, and four words are deleted

   About 30% of the Foundation’s roughly 120 staff are laid off, concentrated in advocacy and global programs. Four months later Firefox publishes terms of use for the first time, and a line is removed from the FAQ. The old answer to whether Firefox sells your data had been: “Nope. Never have, never will.” Mozilla explains that blanket claims are no longer safe because the legal definition of a “sale” of data has broadened.

6. 1 December 2026

   The current agreement expires

   The December 2025 final judgment in the Google case declines to ban default-search payments but caps any such deal with a browser developer at one year. Cross-appeals are pending. What replaces the arrangement is unknown as I write this, which is the honest place to stop.

Look at what the record actually shows and what it does not. It does not show anyone abandoning the manifesto — the manifesto got longer. It does not show a decision to depend on one customer; it shows a decade of individually correct renewals. It does not show anyone deciding to become an advertising business; it shows an acquisition framed, sincerely, in the vocabulary of privacy.

What it shows is an entity acting freely, over twenty-three years, each action leaving an artifact, the artifacts accumulating into a history, and the history training every prediction machine in range — including the institution’s own — to expect a particular kind of organisation. And somewhere in there, the thing the whole structure was built to keep dominant stopped being the thing that dominated.

Four words got deleted from an FAQ in February 2025, and that is the closest this entire story comes to a moment. It is not one. It is an artifact, like the other thousand, and the only reason it is legible is that it was a sentence rather than a renewal.

07Boeing, briefly and carefully

A second case, because one is an anecdote — and I will be exact about what is established and what is merely widely said, since this is the territory where careless writing becomes defamatory.

Boeing completed its merger with McDonnell Douglas on 1 August 1997. In 2001 it moved its headquarters from Seattle to Chicago, and the announcement said plainly that the headquarters needed to be central to its operating units, customers and the financial community — “but separate from our existing operations.” That is Boeing’s own language, in Boeing’s own release, and it is a decision about where an institution’s attention would live.

The widely repeated claim that the merger converted an engineering culture into a finance culture is not an official finding. No government report concludes it. The congressional report discusses it explicitly as something “widely reported” and attributed to “many observers,” with a magazine in the footnote. Harry Stonecipher’s remark that the intent was for Boeing to be “run like a business rather than a great engineering firm” is genuine and appears in that report, sourced to a 2004 newspaper interview. Treat the culture thesis as journalism and economics, not as adjudicated fact.

What is in the record with numbers attached: between 2013 and 2019 Boeing spent $17.4 billion on dividends and $43.1 billion on share buybacks — a finding by the economist William Lazonick, cited approvingly in the congressional report. Sixty billion dollars of artifacts, each one lawful, each one approved, each one teaching every prediction machine in the vicinity what kind of company this now was.

On the 737 MAX I will state only what was legally established. Boeing as a corporation entered a deferred prosecution agreement in January 2021 for conspiracy to defraud the United States, concerning the deception of an FAA evaluation group about MCAS, and admitted the statement of facts was true. The matter concluded in 2025 under a non-prosecution agreement; Boeing has never been convicted of any MAX-related crime, and the only individual ever charged was acquitted on all counts in March 2022. The congressional finding is that Boeing made faulty assumptions about MCAS and withheld crucial information, including concealing the existence of MCAS from pilots.

I am not telling you the buybacks caused the crashes. I have no basis for that and neither does anyone else. I am telling you that an institution’s actions are its artifacts, its artifacts are its history, and its history is what teaches everyone — regulators, airlines, engineers, its own board — what to expect from it. That is a claim about denominators, not about causation in an accident report.

08Why the repair everyone reaches for cannot work

So an institution notices. Usually because something in the magnitude finally moves — a loss, a departure, a story — and for the first time the drift produces a jolt rather than a turn.

And it does, without exception, the one thing that cannot possibly help. It restates its values. It convenes a group, revisits the mission, publishes a refreshed statement of purpose, and hangs the new version where the old version hung.

Look at what that is, in our terms. It is an attempt to reach into the denominator and adjust the imaginary component directly. To turn the angle back by declaring it turned. That is the forbidden operation, and it is forbidden because it is impossible — no entity, however sincere, however well governed, has ever had access to its own E.

A mission statement is not an Actual. It is a description of an intended Actual, which is the same category error as fear and hope: fiction inserted where the record goes. Nobody’s prediction machine is retrained by it, including the machines of the people who wrote it, who will go back to their desks and do tomorrow what the structure makes available tomorrow.

The road out was made of artifacts. The road back is made of artifacts. There is no other material, and the return trip is not shorter.

What would actually work is unglamorous to the point of being unsellable. Change what actions are available, so that a different class of artifact starts entering the record, and then wait — years, plausibly as many years as the drift took — while prediction slowly retrains on the new history. That is the whole method. It cannot be announced, because announcing is not acting. It shows no result in the first year by construction, because P is trained on accumulation.

Which is why institutions almost never do it. The honest repair is indistinguishable from inaction for exactly as long as the people who authorised it remain in their posts.

09What to watch, including at Blender

You cannot feel a rotation, but you can read one — not in the entity’s statements, which are worthless for this, but in the shape of what it produces. Five things to look at, and I mean look, not ask.

1. Concentration of the source

   What proportion of the entity’s inputs arrives from one counterparty? Not because a large funder is sinister, but because dependence narrows the set of available actions, and the available actions are what the record gets made of.

2. The composition of the artifacts

   Line up a year of what the institution actually produced. How much of it is the thing the charter names, and how much is the thing that sustains the thing? Count outputs, not intentions.

3. Which department gets cut first

   A budget is a ranked list of what an entity believes it cannot do without, written under pressure, which makes it more honest than any statement. Watch what goes when something has to go.

4. Language quietly leaving the record

   Not new language added — that is cheap. Old commitments disappearing, and the reasons given, which are usually true and usually legal and usually exactly as significant as they appear.

5. Whether the impossible act is still impossible

   The strongest charters do not exhort. They remove options. If the things that were once structurally unavailable have become merely discouraged, the rotation has already happened and the rest is bookkeeping.

Now turn all five on the case I gave you last week, because a framework you only point at other people is not a framework.

The Blender Foundation would be drifting if its corporate funding concentrated into one or two patrons rather than forty-seven; if the release notes filled with features serving those patrons’ hardware rather than its users’ work; if the first cut in a hard year fell on documentation and community rather than anywhere else; if the words about being free forever quietly acquired qualifications; and above all if the share capital that cannot exist ever became something the board found itself discussing. The last one is the real tell, because it is the only item on the list that is currently impossible.

I am not predicting any of that. I am saying what would have to be true, in advance, so that the claim I made last week can be checked by someone who does not take my word for it — which is the only kind of claim worth making about an institution.

10The claim, and where it breaks

Load-bearing

The impossibility rule does the work and does not depend on anything contestable: no entity touches either term, so every change in a denominator has to have arrived through action, artifact, and record. That is not a claim about corruption; it is the structure of the equation. Corruption is just the name for the case where the direction of travel runs away from the founding idea.

Convention

Calling drift “corruption” is a choice, and an argumentative one. The word normally implies a wrong, and the thing I have described involves no wrong at any step. I keep the word because the outcome is the outcome the word exists to name — an institution that no longer does what it was for — and because reserving it for the criminal case is exactly how the ordinary case escapes notice.

Where the shorthand breaks

“Corruption is a rotation” is precise only if a rotation preserves magnitude, and drift usually changes the institution’s scale as well as its direction. So the honest statement is that the rotation is the part nobody feels, not that nothing else happens. And the sharper problem, which I will not paper over: on the complex log, a larger Ideal makes |E| larger and therefore makes felt surprise smaller — so on this arithmetic idea dampens and never amplifies. Either Actual carries a direction too, or a single complex expectation is the wrong model. I do not know which, and this lesson sits on the unresolved side of it.

Where I would push back on myself

Three places. First, this is unfalsifiable as I have stated it — no one has measured an institution’s θ, and a framework that explains every organisational decline without predicting one is decoration. Section nine is my attempt to owe something checkable; judge it on that. Second, the empirical literature is less friendly to me than I would like: the best panel study of microfinance institutions found no industry-wide mission drift at all, which is a real result cutting against the popular version of my claim. Third, I have chosen two famous cases, and famous cases are selected for having gone wrong. The thousands of organisations that held their angle for a century do not get written about, and I have no idea how many they are.

One last thing, and then you can go.

The reason this lesson follows the last one is that they are the same argument with the sign flipped. A charter is a constraint on available actions. Constrain the actions and you constrain the history; constrain the history and you constrain which Actuals ever arrive to do the retraining. That is how a denominator gets held at an angle for twenty years, and it is the only way, because nobody can hold it there directly.

Loosen the constraint and nothing dramatic happens. Nothing dramatic is what happens. The actions widen, the record diversifies, prediction learns the new behaviour because prediction is honest and has no opinions, and the idea that used to dominate is still written down, still true, still framed on the wall, and no longer the thing the denominator points at.

Nobody stole it. It turned.