Blockchain and Distributed Ledger: The Structure and the Network Types

A distributed ledger is one record of the same facts held by several parties at once, kept in step by an agreed procedure rather than by one party sending copies to the others. A blockchain is one way of building such a record and not another name for it. The structure fixes exactly one thing: no single holder can quietly change an entry everybody else already holds.

Every question worth asking about a shared record turns out to be a question about who holds it and who may write to it, and not a question about software. Naming the holders first settles the rest: which network type this is, what it costs to agree a single entry, what the arrangement takes out of somebody's working day, and whether it was worth building at all. With the holders named wrongly, no amount of engineering rescues the idea. Parties decide more than machinery does.

What is a distributed ledger, and how is it different from an ordinary database?

A wedding fund run by five people is the whole problem in miniature, so start there. Five cousins collect contributions for a cousin's wedding and each of them keeps a small notebook. Every notebook is honest. Each cousin writes down only what passed through their own hands, and one of them writes in pencil, so every notebook is also incomplete. Two weeks before the wedding they sit down to work out what has actually come in, and the evening goes on rather longer than any of them expected. Nobody has done anything wrong. There are simply five records of one set of facts and no agreed way of putting them together.

Now change one thing. Put a single register on the table in the middle, let all five write into it in front of each other, and let none of them take it home. The facts have not changed and the money has not changed. The agreeing now happens at the moment each entry is made rather than two weeks later, so the evening's work at the end has gone.

The single register in the middle is what a distributed ledger is, and the definition is about custody and permission rather than about technology. A distributed ledgerOne record of the same facts held by several parties at once and kept in step by an agreed procedure. is one record of the same facts held by several parties at once, kept in step by a procedure all of them accepted in advance. An ordinary database is the other arrangement: one party holds the record that counts, and everybody else gets told what is in it. If the holder edits a row at four in the afternoon, the other parties find out when they are told, or when they next ask, or never.

A distributed ledger has no centre, and the absence matters as much as anything it has. No holder sits in the middle with arrows radiating out to the rest. Sumeru Bank Limited, an invented bank, is one of five holders and has no more standing in the record than the other four. If Sumeru rewrites entry 0413 on its own machine tonight, the other four still hold the original, and the disagreement is visible the moment anyone looks. Nothing in that protection is cryptographic. The protection is four other people in the room.

How does a blockchain actually construct one of those records?

By grouping entries and then tying each group to the one before it. Entries are collected into a group, the group is closed, and a short value is computed from everything inside it. The resulting value is a fingerprintA short value computed from an entry or a group of entries, which changes if anything inside it changes.: change so much as a digit inside the group and the value that comes out is different. The next group then carries the previous group's fingerprint inside itself. Its own fingerprint therefore depends on the one before, and that one on the group before it, all the way back.

The school register does the same job without any of the vocabulary. In a register where each new sheet begins by copying a short code printed at the foot of the previous sheet, and where the code at the foot of each sheet is worked out from everything written on it, one attendance mark on the second sheet cannot be altered on its own. The code at the foot of the second sheet no longer comes out right, so the code copied onto the third sheet is wrong, and the fourth and the fifth are wrong after it. The change itself has not been prevented. Making the change quietly has been prevented.

A blockchainOne way of constructing a shared record, in which entries are grouped and each group carries a fingerprint of the group before it. makes an alteration visible rather than impossible, and the whole of its reputation rests on that one distinction. The chaining does not put a lock on the past. The chaining removes the possibility of editing the past without anyone noticing. For a record held by several parties, that is the property that actually matters.

Why are the two words not interchangeable?

Blockchain vs Distributed Ledger Technology, and what gets lost when the two words are swapped

A distributed ledger is the arrangement: several parties, one record, an agreed procedure for adding to it. A blockchain is one construction of that record, the one that groups entries and chains the groups together with fingerprints. Every blockchain is therefore a way of keeping a shared record. Not every shared record is a blockchain. A set of parties can agree to keep one record in step by other means entirely, and several do.

The relation is that of a building to a construction method. A load bearing wall is one way of holding a roof up. A frame is another. Somebody who uses the words wall and building as synonyms will end up unable to ask whether the roof needs holding up at all. Whether the roof needs holding up is the only question that decides anything.

The arrangement is what gets lost when the two words are used as one, and the arrangement is the part that decides whether any of this is worth doing. A room asked whether it needs a blockchain produces a debate about software. The same room asked who holds the record today, who is allowed to write to it, and how much time those parties currently spend agreeing their versions with each other settles within ten minutes into either a real case or none at all. Almost every proposal that fails, fails on the second set of questions and never reaches the first.

Who may read and who may write, and how many kinds of network does that give?

Blockchain Network Types, the four of them, numbered

Two questions settle it. Who may read the record, and who may write to it. Each question has two answers, anyone or only named parties, so there are four combinations, and every other difference that gets talked about follows from where a given arrangement sits in that grid. The four are numbered so that a later question can point at one without ambiguity, and the numbering is a teaching device rather than anybody's official list.

Type 3 is the one with a name of its own in ordinary use. A permissionedA record where the parties who may read and write are named in advance under an agreement. arrangement is one where the parties who may read and write are named in advance, and the naming is done in a contract rather than in software. The most important part of a type 3 arrangement is therefore a document that a lawyer drafted, not a component that an engineer built.

Which of the four does a regulated institution actually use?

Type 3, and the reason has almost nothing to do with speed or cost. On any given day a regulated lender has to be able to answer four questions. Who holds a record containing its customers' details? In what country does that holder sit? If that holder walks away, what happens to the record? And who is accountable when something in it turns out to be wrong? On a type 1 or type 2 arrangement the set of holders is whoever turned up, and the four questions have no answers at all. None of that is a criticism of the open types. A named accountable party is a requirement of the institution, not a feature of the record.

The proposal a bank actually receives most often is type 4, a shared record it will hold entirely by itself, and type 4 is an ordinary database carrying the cost of a structure whose whole point is other holders. There is nobody to disagree with an entry, nobody holding a copy that still matches when a row is edited, and nothing to keep in step. Every property that made the arrangement worth discussing came from the other four parties in the picture, and the proposal has removed all four of them.

What is a smart contract, and is it either of those two words?

Smart Contract, and why it is neither smart nor a contract

A smart contractA written rule stored on the shared record that carries itself out when a stated condition is met. is a written rule stored on the shared record that carries itself out when a stated condition is met. The definition is that short. Somebody wrote the wording. The wording sits where every holder can see it. When the condition it names becomes true, the rule runs, and it runs because running is what it does rather than because anyone decided in the moment that it should.

Most people already live with one. A standing instruction at a bank that moves Rs 5,000/- to a landlord on the fifth of every month is exactly this shape: a written rule, held somewhere neutral, that executes on a condition. The instruction will pay the landlord on the fifth of the month after the tenant moved out, and it will pay them while the tenant is asleep, and it will pay them while the tenant is arguing with them. Nobody would call that instruction clever, and nobody would take it to a court as the tenancy agreement. The instruction is automation with the word contract stuck to it.

The property that makes a stored rule attractive and the property that makes it unforgiving are one property. The wording therefore has to be settled before it is stored rather than after. People choose this arrangement because the rule cannot be quietly rewritten by whichever party it happens to favour. The same sentence says the rule cannot be quietly stopped by whichever party it happens to hurt. One half cannot be kept without the other.

What does the structure actually fix?

One thing, and it fits in a sentence. No single holder can quietly change an entry that everybody else already holds. The word doing the work is quietly. Read the sentence twice. A holder can absolutely change its own version. The other holders still hold what was there before, and the chaining makes the alteration announce itself, so no holder can change its own version and have the change pass unnoticed.

Everything a shared record is famous for reduces to that one sentence, and a proposal that promises anything beyond it is promising something the structure does not supply. It does not make the entries true. The structure does not make the parties honest. Nothing in it verifies that the person who sent an instruction was entitled to send it. The structure removes one specific possibility from the world, the possibility of a silent rewrite by one party, and that possibility is worth removing precisely when several parties would otherwise have to trust one of their number to hold the only copy.

What does the structure leave exactly as it was?

Whether the entry was right when it was written. A shared entry proves that all the holders hold the same words. The entry does not prove that the words were true, that the person who sent them meant to send them, or that they were entitled to send them at all. The record starts working after the instruction arrives rather than before, and those three separate questions get no answer from it.

A shared record meets the consent question here, and comes off worse than people expect. A record is provable when it holds who did what, when, on what wording and under what authority. A shared record supplies the when and the what with great confidence, and it is silent about the who and the under what authority unless somebody wrote those into the entry as well. The structure guards the record against later editing and offers nothing at all about what went into it, so a wrong entry becomes a permanent wrong entry rather than a temporary one. That is not a defect. The behaviour is what the structure was built for, applied to a case it was never built to help.

What happens when something is sent wrongly on a record nobody can alter?

The remedy is a second entry that cancels the first, and both of them stay in the record forever. Anyone who has kept a cash book in ink knows the move already. The wrong figure is not scratched out. A scratched out figure is exactly what a scratched out figure looks like to whoever reads the book next. A reversing line goes underneath, and the book now carries the mistake and the correction as two permanent facts. An offsetA second entry that cancels the effect of a first one, because the first cannot be removed. is that reversing line, and on a shared record it is the only remedy available.

Sumeru Bank Limited found out what that costs in ordinary counting. Its pilot carried 1,200 instructions on a permissioned distributed ledger shared with 4 other institutions. Three of the 1,200 were sent wrongly, or 0.25 per cent. Not one of the three could be removed. Each needed a second instruction sending the amount back, so the record ended the pilot holding 1,203 entries describing 1,197 intended transfers. The instruction count and the entry count separated the moment something went wrong and they never came back together.

Six entries now describe three mistakes, and they will still be there in ten years. At 0.25 per cent that is not a scandal. At a rate a person would call ordinary it becomes one. On a record like this a correction never replaces anything. The correction accumulates beside the mistake, and anyone reading the record later has to work out which pairs cancel and which do not.

Where does a shared record save real work, and where does it save none at all?

Here is the test, and it costs nothing to apply. Count the parties who each keep their own record of the same events. Then count the pairs among them. A pair is where the work is: two parties, two records, one conversation to agree them. Each such conversation is a reconciliationChecking two records of the same events against each other, which needs two parties to be worth doing., and the pairs are the only thing a shared record actually removes. The pairs among a set of parties come to the number of parties times one less than that, all halved. Five parties make ten pairs. Eight make twenty eight. One party makes none. There is nobody on the other side of the table.

So the decision has two questions in it and neither one is about technology. Are there several parties who each keep their own record of the same events? And do those parties currently spend real time agreeing those records with each other? If either answer is no, a shared record has no work to do in that situation, whatever else is true about it.

What did the pilot at Sumeru Bank Limited actually carry?

The pilot ran on a tokenised deposit, meaning a claim on the bank recorded as a token on the shared record, over a permissioned arrangement shared with 4 other institutions. The deposit sat on network type 3 in the numbering above. SettlementThe moment a transfer becomes final between the parties to it. between the five fell from 1 working day to under a minute. The fall is a real result and the reason the pilot is worth reading at all.

The two ends of the table reconcile in both directions and it is worth doing the arithmetic yourself. Take the 1,200 sent, remove the 3 that were wrong, and 1,197 transfers were actually intended. Take the same 1,200 and add the 3 offsets that had to be written afterwards, and the record holds 1,203. The 6 entries between those two numbers are three mistakes and three corrections, and they are the permanent residue of the 0.25 per cent. The pilot did not go to production at Sumeru Bank Limited.