June 2016. An attacker exploits a reentrancy flaw in a smart contract called The DAO and drains about 3.6 million ETH.
The response was not a patch. It was an irregular state change: a hard fork at block 1,920,000 on 20 July 2016 that reached into the ledger and moved the funds to a recovery contract. Not a new rule going forward. A rewriting of what the record said had happened.
It was ratified by a polling tool called Carbonvote, built hastily for the purpose, weighted one vote per ether, open for under 24 hours.
The numbers are worth reading slowly. Of 82,054,716 ETH in existence, 4,542,416 voted. Turnout: 5.5 percent. Of those, 3,964,516 voted in favour and 577,899 against. Roughly a quarter of the yes votes came from a single address.
A minority refused and kept running the unforked chain. It is still running. It is called Ethereum Classic, and it is the same chain that gets 51 percent attacked in F101-02, which is its own kind of commentary.
The decision point is not whether the fork was right. People still argue about that and this lesson takes no side.
It is that the property everyone believed they had bought was nobody can change the record. When enough of the people who mattered wanted it changed, it changed, in about a month.
Decentralization is the most used and least examined word in this industry. It is applied to systems that are not decentralized in any meaningful sense, and withheld from systems that are, and it is almost never accompanied by a statement of what it is supposed to deliver.
So let us be narrow about it.
It buys exactly one thing
No single party can change the record on their own.
That is the property. It is genuinely new, it is worth a great deal, and it is much smaller than what gets claimed.
Notice what is not on that list. It does not make anything faster; a global replicated log is far slower than one company's database. It does not make anything cheaper. It does not make it private, since most chains are radically public. It does not make the code correct, as The DAO demonstrated at some expense. It does not make you safe, which is the entire Freshman level.
Everything else attributed to decentralization is either downstream of that one property or is marketing.
The four costs
Speed. Every participant validating everything means the system runs at the pace of broad agreement. This is the tradeoff behind every chain that markets itself on throughput: something was given up, and it is usually how many independent parties can afford to keep up.
Cost. Thousands of machines doing the same computation is expensive by construction. You are paying for redundancy, and redundancy is the product.
Irreversibility, which cuts both ways. No operator to reverse fraud is also no operator to reverse your mistake. F101-01's autopsy is $520,000 gone with nothing malfunctioning, and it came back only because a company chose to be generous.
Coordination. Changing anything requires convincing a distributed group of strangers with divergent interests. This is a defence when someone wants a change you would hate, and a liability when something needs fixing quickly.
That last cost is the one people underestimate, because it is also the mechanism by which immutability actually works.
Immutability is a social fact with a price tag
The common belief is that a blockchain cannot be changed because mathematics forbids it.
That is not the case, and the DAO is the proof. What the mathematics guarantees is that changing history is detectable and requires everyone to agree to run different software. The seal on block 47 does not stop you editing it; it guarantees that everyone will notice.
So immutability is a coordination cost, not a physical law. It holds because getting thousands of independent parties to accept a rewrite is hard, expensive and visible. Raise the stakes enough and it can be paid, and in 2016 it was.
Here is the part that redeems the story rather than undermining it. The rewrite could not be done quietly. It required public argument, published code, and every participant deciding individually. And when a minority refused, they simply kept running the old chain, which still exists.
That is what the guarantee actually is. Not that the record can never change, but that changing it requires open coordination and leaves a permanent receipt in the form of a second chain. That is narrower than the advertised promise and it is still, historically, unprecedented.
"87 percent voted in favour" is the figure usually quoted about the DAO fork. Convert it into a share of the whole, which is the only form worth acting on.
Turnout:
4,542,416 / 82,054,716 = 0.0554 = 5.5 percent of all ETH voted at all
Share of total supply voting in favour:
3,964,516 / 82,054,716 = 0.0483 = about 4.8 percent
So "87 percent approved" describes a decision in which roughly 19 out of every 20 units of the currency did not participate.
Now the concentration. About a quarter of the yes votes came from one address:
3,964,516 x 0.25 = about 991,129 ETH from a single holder
Compare that to the entire opposition:
991,129 / 577,899 = 1.71
One address outvoted every objector combined, by more than seventy percent.
None of this establishes that the fork was wrong, and it does not mean the process was illegitimate; there was no better tool available and the deadline was real. What it establishes is what the number meant, which is different from what the number sounded like.
Carry the method rather than the case. Whenever you meet a governance claim, from a protocol vote, a DAO proposal, a token holder decision, ask three questions: what share of the whole participated, how concentrated were the votes that did, and how long was the window? Those three convert almost every impressive governance percentage into something much more ordinary, and S202 and J-level governance material assume you can do it.
Finding the chokepoints
Since the label tells you nothing, assess a system by asking where it can be squeezed. Four questions, in order of how often they find something.
Who can halt it? Is there a pause function, an admin key, an upgrade multisig? Many things described as decentralized have a small group who can stop them, and that is not a scandal by itself, but it is the answer to the question.
Who can upgrade it? Contract upgradability is a chokepoint with excellent intentions. If a handful of keys can change what the code does tomorrow, today's audit describes today's code only.
Who runs the infrastructure everyone actually uses? This is the one people miss, and F101-04 is entirely about it. A perfectly decentralized chain reached by everyone through two API providers has a real-world chokepoint that its consensus mechanism knows nothing about.
Who holds enough of the token to decide alone? The DAO's answer was one address holding more than the entire opposition.
A system that fails one of these is not fraudulent. It is just less decentralized than the word implies, and knowing which parts are load-bearing is worth more than the label.
Real decentralization means nobody is in charge, so a project with a leader is fake.
This confuses two different things and produces bad judgment in both directions.
Decentralization is a property of the protocol: whether one party can change the record alone. It is not a property of the organisation around it. Bitcoin has maintainers. Ethereum has a foundation, a research group and named client teams. Both are decentralized in the sense that matters, because none of those people can alter your balance.
Conversely, a project with no visible leader is not thereby decentralized. Anonymous teams have shipped upgradable contracts with a single admin key many times, and the absence of a name in front of it changed nothing about who could take the funds.
The useful question is never who is in charge of the project. It is which specific powers exist, and who holds them. That question has a factual answer you can go and check, and the label does not.
The honest summary
Decentralization is a purchase with a price. You get one property, and you pay in speed, cost, irreversibility and coordination difficulty.
For a lot of uses that is a bad trade, and being able to say so is part of understanding it. A company's internal database should not be a blockchain. For the specific case of holding value that nobody can confiscate or inflate away without open coordination, it is the only mechanism anyone has built that works.
Knowing which case you are in is the skill. F102 is where we ask whether the trade is worth it at all.
Decentralization buys exactly one property, that no single party can change the record alone, and charges for it in speed, cost, irreversibility and coordination. Immutability is therefore a social commitment backed by coordination cost rather than a law, which the DAO fork demonstrated by rewriting the ledger in about a month. What survived is narrower and real: the rewrite could not be done quietly, and the minority who refused still run the original chain. Judge any system by its chokepoints rather than its label, and convert every governance percentage into a share of the whole, because 87 percent in favour meant 4.8 percent of the supply and one address outvoting all objectors combined.