Consensus Is Pressure
I have spent a large part of my working life on machines that have to agree with each other, and the thing nobody warns you about at the beginning is that consensus has almost nothing to do with truth.
It is the promise that separate machines can share one story. Each node keeps its own clock, its own copy, its own damaged view of the network. Messages arrive late. Packets vanish. A machine goes dark and comes back claiming nothing happened. The system still needs an answer, and the answer it produces is not the true version of events. It is the version that enough machines accepted.
I found that unsettling for years before I worked out that it was a plot.
Because it is exactly the room where everyone knows what happened and nobody can afford to say it first. A blockchain does not ask who deserves to be believed. It asks which account of history survived enough checks to be written down, which is a question I have watched humans answer in meetings, badly, with the same result.
A ledger feels clean because it looks final. Blocks stack, hashes link, transactions lock into place, and the machine gives a strong impression of moral order. Inputs go in, rules apply, outputs arrive. People are what bring the disorder. They arrive carrying fear and debt and pride and private bargains, and the old human habit of moving blame until it settles on the weakest hand. Put those people next to a system that preserves evidence by design and you barely have to invent anything.
A commit becomes an alibi. A signature becomes a weapon. A validator becomes a hostage.
The technical question in all of this is simple. Can the network agree? The question I actually care about is worse, and it is who benefits from the agreement.
None of which is because blockchains are magic. They are not, and I would not have written this book if I thought they were. They are brittle, expensive, ingenious systems with clear limits and genuinely ugly failure modes, and they run on incentives, on keys, and on people behaving predictably under stress. People rarely behave predictably under stress. That is not a flaw in the design. It is the part of the design that was always going to be someone else's problem.
The failure mode I like best is the partition. The room splits. Half the network sees one truth and half sees another, and each side can sound entirely rational, and each side can point at logs and timestamps and signatures and rules. The argument gets technical because everyone involved needs it to look less personal. Then the network heals, and the system has to decide which past survives.
That is thriller material, and it never needed a chase. It needs a delay. A missing signature. A block that should have landed and did not. Someone watching a cursor blink while money and trust and blood move somewhere off the screen.
The machine will not care which way it goes. I spent a long time thinking that was the frightening part of writing this, and I have changed my mind. The frightening part is how quickly the people around it learn to hide behind that.