What a chain actually is

A chain is a data structure. It is a list, with one unusual rule about what each element has to contain, and the rule is enforced by arithmetic rather than by permissions. That is genuinely clever, and it is also completely mundane once seen, in the way that double entry bookkeeping is clever and mundane. Nothing in it is magic, nothing in it is new mathematics, and nothing in it knows whether the entry it is protecting is true.

The claim worth making about it is narrow and strong: you can hand the whole register to someone who does not trust you, and they can check for themselves that it has not been altered since it was written. The claims worth refusing are the wide ones, and refusing them is easier once the structure fits in one diagram.

• Hash. A function taking any input and returning a fixed length output. Also called a digest or a fingerprint. Part 1 tonight, and practitioner session 1.
• SHA-256. One particular hash function, the one Harbourgate uses. It returns 64 hexadecimal characters no matter what goes in.
• Block. One entry in the register, packaged with its position, its timestamp, the previous block's hash, and a number called a nonce.
• Genesis block. The first block, which has no predecessor to point at. Part 2 tonight.
• Avalanche. The property that a tiny change to the input produces an unrecognisably different digest. Without it the rest does not work.
• Nonce. A number a block carries that can be varied freely without changing what the block says. Idle at difficulty 0, and the whole subject of session 3.
• Merkle root. One hash standing for a whole set of records. Practitioner session 6.
• Key management. Who holds which secret, and how it is rotated and revoked. A discipline of its own, covered in the security courses rather than here.

Imagine a fingerprint that behaved sensibly: similar inputs, similar outputs. Anyone wanting to alter an entry would simply nudge it until the digest landed back where it started, in the way you close in on a target by getting warmer. The whole structure would be worthless, because the attacker would have a gradient to climb.

The avalanche property removes the gradient. Every attempt is a fresh draw with no information carried over from the last one, so there is no warmer and no colder, only right or wrong. That is why a hash function is described as one way. Running it forwards is instant. Running it backwards has no method better than trying inputs.

A block is hashed as a whole, not as a line of text. Its position in the register, its timestamp, its entry, the previous block's hash and its nonce all go in together. Change any one of those five and the digest changes, which means the digest is a statement about all of them at once.

This matters for what you are allowed to claim. A matching hash says the block is byte for byte what it was when the hash was taken. It says nothing about whether the inspector actually looked at the seal.

There is a long running argument about whether a chain is a new idea at all. One side points out that hash linked records were published in the late 1970s and used in practice through the 1990s, that a list in which each element carries its predecessor's digest is a textbook structure, and that calling it a breakthrough is a sales pitch attached to a linked list.

The other side says the structure was never the contribution. The contribution was the packaging: a way for parties who do not trust each other to hold copies of the same register and agree on which one is current, with no administrator in the middle. On that view the data structure is old and the system around it is not.

Both readings are held by serious people and they lead to different purchasing decisions, so it is worth knowing which one the person opposite you holds. This session is deliberately about the part nobody disputes.

Block 0 is the exception that shows what the rule is for. Every other block inherits its credibility from the one before it, and block 0 has no predecessor, so its previous hash field is filled with a row of zeros standing in for a block that does not exist.

The consequence is worth saying plainly. The structure carries confidence forward from the start of the register, and it cannot manufacture any at the start. Whoever wrote block 0 and whoever was watching are the foundation, and no amount of hashing afterwards improves that.

Hand a copy of the Harbourgate register to a buyer who thinks the port authority is covering something up. They recompute all six hashes themselves and compare each block's previous hash field with the hash of the block above it. If everything matches, they know the six entries are exactly the six entries that were written, in that order.

That is a real promise and it is narrower than it sounds. It covers the entries as written, not the events they describe, and it says nothing about entries that were never written down at all.

Harbourgate's blocks carry fixed timestamps starting at 1758000000 and advancing 600 seconds per block, which makes the register reproducible: anyone rebuilding it from the six entries gets the same six hashes rather than six new ones. A real register takes its timestamps from a clock, and whose clock that is turns out to be a harder question than it looks.

The nonce field sits in every block doing nothing at difficulty 0. It exists so that a block's hash can be changed without changing what the block says, which sounds like a flaw and is the foundation of session 3.