"Decentralized" is the most abused word in this industry, and the abuse is almost always the same: the claim is about the protocol, and the reader assumes it is about the power.
A network can be architecturally decentralized and operationally concentrated at the same time. Bitcoin is the clearest example: no one controls the rules, and a few pools produce most of the blocks while a shrinking number of operators hold the archive. The protocol is decentralized. The practice is drifting.
The number that decides which of those is true is not hashrate. It is how many people can check the books themselves.
What a node actually does A node is a computer that downloads the chain, validates every rule, and refuses anything that breaks them.
Miners produce blocks. Nodes decide whether to accept them. If a miner produces a block that pays itself too much, or double- spends, or violates the emission schedule, every honest node rejects it and the miner has wasted electricity. That is the entire enforcement mechanism of a proof-of-work currency, and it only works if enough independent people are running the software to make rejection meaningful.
This is why chain size matters more than almost any other technical metric. A chain that requires a data-center-grade machine to validate is a chain whose rules are enforced by the sort of operator who can afford one. A chain that fits on a laptop is a chain whose rules are enforced by anyone who cares.
The arithmetic Epic's complete archival history — every block since September 2019, more than three million of them — occupies approximately four gigabytes.
Bitcoin's is roughly 750 gigabytes. A pruned Bitcoin node can run on seven to ten gigabytes of disk, but only after downloading and verifying all 750 first, and the archive it depends on stays roughly two hundred times larger than Epic's entire history.
Monero's full chain is around two hundred gigabytes. Zcash's node data is in the same range, having passed three hundred.
Four gigabytes is the size of a photo library. It downloads over a domestic connection in an afternoon, syncs on a laptop that was mid-range five years ago, and lives on a disk you already own.
This is not an engineering achievement so much as a property of the ledger format. Bitcoin stores every transaction forever because the history is the proof. Mimblewimble deletes what no longer matters: when an output is spent, the output and the input that consumed it cancel and are removed, leaving a small kernel that proves a valid transaction happened without saying anything about it.
Seven years and three million blocks later, the result is measured rather than projected. That is the difference between a scalability claim and a scalability fact.
What you can still check Deleting history raises an obvious question: if the chain forgets, what stops someone from inventing money?
The answer is in the kernels and the commitments. Every node can verify that the sum of all outputs equals the sum of all inputs plus the scheduled emission — that the money supply is exactly what the protocol says it should be at this height. It can verify that no output is spent twice, because the set of unspent outputs is complete even though the spent ones are gone. It can verify every signature.
What it cannot do is reconstruct who paid whom in 2021, because that data no longer exists anywhere. That is not a side effect. It is the privacy property and the size property arriving as one thing.
There is an asymmetry here that the 2026 technical paper states rather than hides: Epic's default sync verifies the full state but skips historical header proof-of-work outside a recent window, where Bitcoin's default replays full validation from genesis. Epic offers a full header-validation mode by configuration. A reader comparing "validation" across chains should know that the defaults are not identical.
The exception the book has to mention Epic's operational record has one interruption, and it belongs in this chapter rather than in a footnote.
In March 2021 an attacker exploited a value-overflow flaw — the same class of bug as Bitcoin's CVE-2010-5139, which in 2010 minted 184 billion counterfeit bitcoin before being caught and rolled back. Epic's response followed the same precedent: patch the flaw, invalidate the exploited history by rolling back to a pre- exploit block, and resume. The chain was paused for several weeks.
Two things should be said about that, and neither cancels the other.
The first is that a rollback is a real centralization event. Someone decided which history would stand, and the network followed — because at that point in its life there was no realistic alternative. A reader who values immutability should treat this as a scar, not a rounding error.
The second is that the alternative was worse: a chain carrying billions of counterfeit coins, which is the outcome Bitcoin itself refused in 2010 for the same reason. Every proof-of-work currency has this choice in its history or in its future. Epic's is public, dated, and documented.
The incentive nobody talks about Running a node does not pay.
There is no reward for validating, no yield, no token. People run nodes because they want to verify their own transactions without trusting someone else, because they run a business that accepts the coin, or because they believe the network should have more independent validators than it currently does.
That means node count is a volunteer number, and volunteer numbers are fragile everywhere. Every small chain has the same structural weakness: miners are paid, node operators are not, so the ratio drifts toward miners. Epic is not exempt from this, and no honest chapter can claim otherwise.
What Epic can claim is that the barrier is low enough that the decision is available to an ordinary person. That is not the same as saying many ordinary people make it. Node connections have been observed from more than 130 countries as of July 2026 — and the technical paper is careful to call that a proxy for geographic reach measured by geolocated IP addresses, not a count of users or a measure of decentralization.
What this is worth The argument of this book is that money should be verifiable by the people who hold it. That argument is either supported by a chain you can download or it is a slogan.
Four gigabytes means a person with a laptop and an evening can verify, independently, that:
The supply is exactly what the protocol says it is. No coin has been spent twice. The block that paid a miner was valid. Nobody changed the rules while they were not looking.
That is what "governed by no one" means in practice. Not that no one is in charge — someone is always trying — but that no one can change the books without every validating machine on the network noticing and refusing.
Running one The practical requirements are modest, and worth stating plainly because most people assume they are not.
A machine with a few gigabytes of free disk, a domestic internet connection, and the patience to let it sync. The software is open source and free. There is no account, no registration, no permission, and nothing to configure beyond a data directory and, if you want it, a Tor proxy.
What you get for that is a node that answers to you. It validates your own transactions before your wallet broadcasts them, it tells you the real state of the chain rather than a block explorer's summary of it, and it participates in the only mechanism that makes "governed by no one" more than a phrase — the collective refusal of the machines that validate to accept a block that breaks the rules.
What you do not get is money. There is no reward, no yield, no incentive program. If that sounds like a weakness, it is: it is the weakness every proof-of-work network shares, and the reason node count is always lower than the design would like.
Everything else in this book rests on that. Which is why the next chapter is about the thing you do with a verified chain: send a payment to another person, and have it arrive in a minute for about a cent.