FTX was the second largest crypto exchange in the world, with a founder on magazine covers and an advertising budget that reached stadium naming rights.
Customer funds had been moved to Alameda Research, an affiliated trading firm. The shortfall was reported at around $8B.
A competitor announced it would sell its holding of FTX's own token. Roughly $5B of withdrawal requests arrived in 72 hours. Withdrawals were halted on 8 November 2022. Bankruptcy followed on 11 November.
Now look at the mechanism, because it is the reason this autopsy opens a lesson about whether any of this is worth anything.
Not one part of it required a blockchain, and not one part of it was prevented by one. Customer balances sat with a custodian, in a private database, moved by internal instruction. This is an ordinary fraud, of a kind that predates computers, and it happened to be denominated in crypto.
The property crypto genuinely offers, that you can hold assets yourself and that reserves can be publicly verified on a ledger anyone can read, was available to every customer and to the company. Neither used it.
The estate later recovered a great deal, and a 2024 plan paid 98 percent of creditors 119 percent of their allowed claims. That number is doing more work than it appears to, and we come back to it below.
This is the lesson where we stop describing the machine and ask whether it should exist.
You will meet two kinds of people on this subject and both are useless. One believes it fixes money, poverty and freedom. The other believes it is entirely fraud with a database attached. Neither has looked closely.
Our position, and you are free to reject it: there is a small set of real claims, a small set of real objections, and most of the argument you have heard is about neither.
Four claims that survive
Censorship-resistant holding. You can hold an asset that no institution can freeze, seize by instruction, or restrict access to. Cyprus is the demonstration. Condition: you actually self-custody. An exchange balance has none of this property, which is the autopsy.
Permissionless access. Anyone with a connection can hold and transact, without an account, an approval, a credit check, or a jurisdiction. Condition: you can get in and out. If your on-ramp and off-ramp both require permission, you have permissionless middle and permissioned ends, which is the honest state of most people's situation.
Verifiable settlement. Ownership and transfer are publicly checkable by anyone, without trusting a statement. Condition: you look. The information that would have shown FTX's problem partly existed, and almost nobody was reading it.
Credible supply commitments. A rule about issuance can be made hard to change, backed by the coordination cost from F101-03 rather than by an institution's promise. Condition: the chain is actually decentralised, which most are not.
Notice what is absent. Not on the list: that it goes up, that it is a good investment, that it replaces banks, that it is faster, or that it is cheaper. Some of those are sometimes true. None is a property of the technology.
Four objections that survive
Most activity is speculation. Not a slur, an observation. The dominant use is trading, and the dominant motivation is price. The useful applications exist and are a minority of the volume.
The industry keeps rebuilding what it set out to replace. Custodians, leverage, opaque balance sheets, related-party lending. FTX was not a crypto failure, it was a bank failure staged on crypto rails, and the industry produces these regularly.
The fraud is enormous and it is not incidental. F106-01's figure: a record $17B estimated in scams and fraud in 2025. Irreversibility, pseudonymity and permissionless access are the same properties on the list above, seen from the other side. The features and the fraud are the same features. That is the honest form of the objection and there is no version of the technology that keeps one without the other.
The claimed use cases are mostly unrealised. Payments are a small share of activity. Most tokenised assets are not held by the people the pitch describes. Much of what is built is infrastructure for more trading.
Any argument for crypto that cannot hold all four of those simultaneously is advocacy, not analysis.
The distinction that resolves most of the argument
Almost every bad conversation about crypto conflates two things.
A property of the technology. What the protocol makes possible. Self-custody. Public verification. Rules that are expensive to change.
A property of the industry. What the companies and the culture built on top of it. Custodians. Leverage. Marketing. Fraud.
FTX was a failure of the second, in a way the first would have prevented. Every customer who self-custodied lost nothing. Every customer who trusted a balance in a company's database was in the position of a Cypriot depositor with fewer protections and no insurance.
That does not exonerate the technology, because a technology whose industry reliably produces these is fairly judged partly on that. But it does tell you which criticism you are making, and it tells you what to do differently, which "crypto is a scam" does not.
The FTX estate's 2024 plan paid 98 percent of creditors 119 percent of their allowed claims. That reads like everyone was made whole and then some. It is standard bankruptcy practice and it cost creditors a great deal, and the reason is denomination.
Claims were valued in dollars, as of the petition date, November 2022. Not in the assets held.
Take a customer with 1 BTC on the platform. Bitcoin was around $16,000 in November 2022.
Allowed claim: $16,000 Paid at 119 percent: 16,000 x 1.19 = $19,040
Now suppose bitcoin is $90,000 when the distribution lands. What does $19,040 buy?
19,040 / 90,000 = 0.212 BTC
They started with 1 BTC. They ended with the ability to repurchase about 0.21 BTC, having been paid 119 percent of their claim. That is a loss of roughly 79 percent of the asset, reported as a recovery of 119 percent.
Both numbers are true. They measure different things, and the units are the entire disagreement.
Generalise it, because this is the transferable part. Recovery measured in units of the asset is:
(1.19 x price at filing) / (price at distribution)
That is at least 1 only while the price at distribution stays below 1.19 times the price at filing. So the 19 percent uplift buys a creditor exactly 19 percent of headroom: if the asset rose by less than that, they came out ahead in asset terms, and above it every further percentage point of appreciation is a straight loss.
Bitcoin at $90,000 against $16,000 is a rise of about 460 percent, which is not close to the 19 percent of cover the plan provided.
Two things follow. First, headline recovery percentages in a bankruptcy tell you almost nothing until you ask what they are denominated in. Second, and more practically: this is the mechanism by which a custodial failure costs you even when the estate does well, and it is not a scandal, it is how insolvency works everywhere. Which is a reason to care about custody before there is a problem, and not an argument that anyone cheated afterwards.
FTX proves crypto does not work.
It proves a custodian failed, which is a thing custodians have done since custody existed.
The test is what a customer would have needed to do to be unaffected, and the answer is available: withdraw to a wallet they controlled. Everyone who did that was untouched, on the same day, in the same market, holding the same assets. That is not a defence of the industry and it is a fact about where the failure was located.
The strong version of the criticism is better and worth taking seriously: if the technology's central promise is that you do not need a trusted third party, why did nearly everyone use one? Because self-custody is difficult, unforgiving and irreversible, and most people, correctly assessing their own competence, chose a custodian.
That is the real indictment, and it is not about FTX. It is that the industry spent a decade building the thing it was supposed to make unnecessary, because the alternative was too hard to use.
It is also, precisely, why this Academy exists and why F103 to F106 are where the real work is.
What we are actually claiming
For the avoidance of doubt, since you should know what a course's position is:
We think self-custody of a small, deliberate portion of your assets is a reasonable thing for a competent adult to do, that the skills are learnable, that most people who lose money do so through the failure modes in F106 rather than through market moves, and that whether any of it appreciates is not something we know or will pretend to.
We are not claiming crypto is a good investment, that it will replace anything, or that you should hold more of it. P10 in our design principles forbids us from implying profitability, and we mean it.
If, having read F102, you conclude the trade is not worth it for you, you have used this course correctly.
Four claims survive scrutiny, and every one is conditional: censorship-resistant holding if you self-custody, permissionless access if both ends are open, verifiable settlement if you look, and credible supply rules if the chain is genuinely decentralised. Four objections survive too, and the sharpest is that the features and the fraud are the same features. Keep separate what the technology makes possible from what the industry built, because FTX was an ordinary custodial fraud that the technology's own property would have prevented and almost nobody used. And when a bankruptcy reports 119 percent recovery, ask what it is denominated in, because in the asset it can be a 79 percent loss.