Double Entry

3.0 What this chapter gives you#

1. You will be able to state the rule of double entry in one sentence — never write down one end of a journey — and explain why a record-keeping method that checks itself survived seven hundred years without replacement.
2. You will be able to define debit and credit correctly as left and right, and explain why the same £3.20 is a debit in Lloyds’ ledger and a credit in the cardholder’s.
3. You will be able to post a VAT-inclusive £3.20 coffee sale as a three-legged entry, and say why the shop’s cash rose by £3.20 while its equity rose by only £2.26.
4. You will be able to explain why “double” refers to the two sides rather than to a count of lines, and why a model hard-coded to one debit and one credit cannot express a partial refund.
5. You will be able to name the errors a trial balance cannot catch — omission, commission, principle, original entry, reversal and compensation — and explain why fraud generally balances beautifully.
6. You will be able to trace one card payment through the issuer’s, acquirer’s, merchant’s and cardholder’s books, and show the same interchange appearing as income in one and cost in another.
7. You will be able to explain why a card-accepting merchant books a receivable rather than cash, and why that gap has to be funded.
8. You will be able to explain why an authorisation posts nothing at all, why “pending” exists on the app, and why a refund is a fresh transaction rather than a cancellation.
9. You will be able to describe what happens at the apex of the system, where an interbank payment merely relabels which reserves account the Bank of England owes.
10. You will be able to list the invariants a ledger implementation must enforce: balance at write time, immutable append-only postings, signed integers in minor units, idempotency keys, multi-leg entries and named accounts for in-flight money.

There is a sentence that people say without noticing what it commits them to. They say: I spent three pounds twenty on a coffee.

Notice how many facts that sentence contains. It sounds like one. It is at least two, and if you are the coffee shop it is four or five. Three pounds twenty left something. Three pounds twenty arrived somewhere. The shop is three pounds twenty richer in one respect and slightly poorer in another, because a cup of coffee that cost it forty-one pence to make no longer exists. Some of that three pounds twenty was never the shop’s money at all; fifty-three pence of it belongs to His Majesty’s Revenue and Customs and the shop is merely holding it. None of this is philosophy. It is all recorded, in books, by people whose job it is, and the discipline they use to record it is seven hundred years old and has not needed replacing.

The discipline is called double entry, and it is the single most load-bearing idea in this book.

Everything that follows — clearing, settlement, netting, interchange, chargebacks, the reason a refund takes five days, the reason your available balance is not your ledger balance, the reason a bank can fail on a Friday and reopen on a Monday with every customer’s number intact — is double entry operating at a scale and speed that would have astonished the Venetians who invented it. Not an analogy for double entry. Not something like double entry. It is double entry, running on faster machinery.

The previous chapter established that your balance is not a substance but a record. This chapter is about the rule that governs how such records may legally and sensibly be changed. Get this chapter and the rest of the book becomes a set of variations. Skip it and every later chapter will feel like a list of arbitrary industry customs.

The plain version#

Every fact about money is a journey, and a journey has two ends.

If your friend hands you five pounds, that is not one thing that happened. It is two things that happened at the same instant: you have five pounds more, and your friend has five pounds less. If you write down only your half, your notebook is claiming that five pounds appeared out of nowhere. Notebooks that claim money appears out of nowhere are, sooner or later, wrong about everything.

So the rule is this: never write down one end of a journey. Always write down both.

That is the whole idea. Everything else is bookkeeping for how to do it tidily.

The two columns#

Draw a line down the middle of a page. On the left you write where the money went to. On the right you write where it came from.

Your friend hands you five pounds:

Two ends. Same amount on both sides. If the two sides do not match, you have made a mistake, and you know it immediately, without anybody having to tell you.

That last part is the reason the idea survived. It is not merely a tidy way to keep records. It is a record-keeping method that checks itself. If you add up everything in the left column and everything in the right column, the two totals must be identical. If they are not, something is wrong and you go and find it. A shopkeeper in 1350 with no calculator, no computer and no auditor could catch his own errors before they compounded. That is an enormous thing to be able to do.

Aisha’s cake stall#

Aisha is thirteen. She is running a cake stall at the school summer fair and she has decided to keep proper books, because her uncle is an accountant and has been unbearable on the subject.

She starts by putting in £20 of her own money, so there is a float. Then she spends £12 on flour, eggs, butter and icing sugar. She spends £3 on greaseproof paper and little labels. The school charges £5 to hire a table. On the day, she sells £34 of cakes.

Here is her book. Every row has a left and a right, and every row balances by itself.

At the end of the fair, how much is in the tin? Add up everything that went into it and take away everything that came out: £20 in, then £12 out, £3 out, £5 out, then £34 in. That is £34.

Now check it from the other direction, which is the point of the whole exercise. The stall took £34 of sales and spent £20 getting there (£12 plus £3 plus £5), so it made £14 of profit. It also still owes Aisha the £20 float she put in. So the total claim on the stall is £20 plus £14, which is £34 — exactly the cash in the tin.

Two completely separate routes to the same number. That is not a coincidence and it is not luck. It is a consequence of never writing down half a journey. Aisha now knows, without checking anything else, that her book is internally consistent. If she had counted £31 in the tin, she would know instantly that either £3 had gone missing or she had forgotten to write something down, and she would go looking. Before double entry, she would simply have had £31 and a vague feeling.

The two words that confuse everybody#

Accountants do not say “left” and “right”. They say debit and credit. Debit means the left-hand column. Credit means the right-hand column. That is all they mean.

Read that again, because almost everyone in Britain believes something else. Almost everyone believes that “debit” means money going out and “credit” means money coming in. That belief comes from bank statements, and bank statements are not lying to you, but they are written from a point of view you have probably never considered.

Look at Aisha’s book. When she sells £34 of cakes and the money goes into the tin, that is a debit to the cash tin. Money coming in is a debit. When she pays £5 for the table and money goes out of the tin, that is a credit to the cash tin. Money going out is a credit.

That is the exact opposite of what your bank statement says, and both are correct. We will resolve this properly in a moment. For now, hold on to the only definition that never changes: debit is the left side, credit is the right side, and every entry needs both.

The coffee, over the counter#

Ama buys a flat white from Kalypso Coffee for £3.20 and pays with a £5 note.

In the shop’s book, that single sale is two rows, not one, because two separate journeys took place.

The first journey is the money. Three pounds twenty arrived in the till, and it arrived from a customer buying something. Left: cash, £3.20. Right: sales, £3.20.

The second journey is the coffee. A cup of coffee left the shop’s stock and went into a customer’s hand. The beans, milk, cup, lid and sleeve cost the shop 41 pence. Left: cost of what was sold, £0.41. Right: stock, £0.41.

Both rows balance. Together they say something a single row could never say: the shop is £3.20 richer in cash, £0.41 poorer in stock, and therefore £2.79 better off overall. A shop that only recorded “sales £3.20” would think it was doing brilliantly right up until the moment it noticed it had no beans.

The same coffee, paid by card#

Now Ama taps her debit card instead.

No money moves. Nothing physical happens at all. What happens is that four different organisations each write down two things.

Ama’s bank writes down that it owes Ama £3.20 less than it did a moment ago, and that it now owes £3.20 to somebody else in the payments chain. Two facts. The shop’s card company writes down that it is owed £3.20 by that chain, and that it owes £3.20 to the shop. Two facts. The shop writes down that it has made a £3.20 sale, and that it is owed £3.20 by its card company. Two facts. And if Ama keeps a household budget, she writes down that she spent £3.20 on coffee and that her bank balance is £3.20 lower. Two facts.

Nobody moved anything. Everybody wrote two things down. That is what a payment is.

And here is the part that makes the whole system hang together: those separate books have to agree with each other. Ama’s bank thinks it now owes £3.20 into the system. Somebody else in the system thinks they are owed £3.20. Those two numbers must match, and at the end of the day somebody checks that they do. When they do not match, that is not a rounding annoyance. That is how fraud is caught, how outages are detected, and how a payments company discovers it has lost real money. There is a whole chapter about it in Volume V and it is titled, without affection, Reconciliation.

Where this came from#

Nobody invented double entry in a single stroke. It grew up among Italian merchants in the thirteenth and fourteenth centuries, in exactly the places you would expect: Florence, Genoa, Venice, and the trading colonies those cities planted around the Mediterranean. The oldest surviving books that unambiguously use the full method belong to a Florentine firm, Giovanni Farolfi & Company, and were written by a partner named Amatino Manucci at the firm’s branch in Salon, in Provence, covering 1299 and 1300. They were identified and analysed by the accounting historian Geoffrey Lee in a 1977 paper in the Accounting Historians Journal. Manucci did not invent the technique. He was simply using it, competently, the way a clerk uses a spreadsheet.

The method spread by apprenticeship for two hundred years before anyone thought to write it down properly for strangers. The man who did was Luca Pacioli, a Franciscan friar and mathematician born around 1447 in Sansepolcro in Tuscany, who died on 19 June 1517. In November 1494 the Venetian printer Paganino de Paganini published Pacioli’s enormous Summa de arithmetica, geometria, proportioni et proportionalità, a general survey of everything then known about arithmetic, algebra and geometry, running to over six hundred pages of dense type.

Buried inside it, as the eleventh treatise of the ninth part, is a section called Particularis de computis et scripturis — “particulars of reckonings and writings”. In it Pacioli sets out what he calls the method of Venice: keep three books, a rough memorandum for jotting things down as they happen, a journal for writing them up in order, and a ledger where each account has its own page with debits on the left and credits on the right. Check the ledger periodically by listing every balance and proving that the two columns agree. He is widely quoted as advising that a merchant should not go to sleep at night until his debits equal his credits.

Pacioli did not invent double entry and never claimed to. What he did was put it on a printing press, at the moment printing presses existed, in the commercial capital of Europe. Three hundred years later Goethe had a character in Wilhelm Meister’s Apprenticeship say of book-keeping by double entry that “it is among the finest inventions of the human mind”. The method has been continuously in use ever since, which makes it older than the Bank of England, older than the pound sterling in its modern form, and older than every institution named anywhere in this book.

Where the plain version stops being true#

The plain version above is a good foundation and it is not a lie. But four things in it will mislead you if you carry them forward unamended, and the fourth one is where fintechs lose money.

Correction one: “debit” does not mean money out#

This is the most common misunderstanding in personal finance and it is entirely the fault of the banking industry, which chose to print the words on statements without explaining whose books they refer to.

Your bank statement says “debit” next to the £3.20 coffee. Aisha’s cake stall book says a debit is money arriving. Both are correct because they are two different books written from two different points of view.

Recall the previous chapter: your deposit is not your money sitting in the bank’s vault. It is the bank’s promise to pay you. In the bank’s own ledger, your current account is a liability — a debt the bank owes you. When you spend £3.20, the bank’s debt to you gets smaller. Reducing a liability is, by the universal convention, a debit. So the bank debits your account, and prints the word on your statement.

Meanwhile, in your own books, your bank account is an asset — something you own. Spending reduces an asset. Reducing an asset is a credit. So in your books, the very same event is a credit.

The same £3.20, at the same instant, is a debit in Lloyds’ ledger and a credit in yours. Neither is wrong. They are mirror images, and they must be mirror images, because the thing that grew smaller for one party is the thing that grew smaller for the other in the opposite direction.

The practical consequence is that “debit” and “credit” are meaningless without two further pieces of information: whose book, and what kind of account. Any sentence of the form “we’ll credit your account” is ambiguous until you know who “we” are. When a merchant says “we will credit your card”, they mean your card balance will move in your favour. When a bank’s core system says “credit the customer account”, it means increase the bank’s liability, which is the same thing. When a payments engineer says “credit the merchant’s payable”, they mean the acquirer now owes the merchant more. Same word, three books, three referents. Systems that blur this produce reversed postings, which are the single most tedious class of bug in financial software.

Correction two: “double” does not mean two lines#

The name is a historical accident and it misleads people who take it literally. “Double” refers to the two sides, not to a count of lines. A single journal entry may have three legs, or twelve, or two hundred. The rule is not “one debit and one credit”. The rule is:

The sum of the debits equals the sum of the credits.

Return to the flat white. The shop is registered for VAT, and hot drinks in the United Kingdom are standard-rated at 20 per cent. Of the £3.20 the customer paid, £2.67 is the shop’s revenue and 53 pence is output tax that the shop is collecting on behalf of HMRC and will hand over later. That is one economic event and it needs three legs:

Debits £3.20, credits £3.20. Balanced, with three lines.

Correspondingly, one real-world event may require several separate journal entries. The coffee sale generates one entry for the money and a second for the movement of stock into cost of sales. The card payment in the technical section below generates one entry when the sale is recorded, another when the acquirer settles, and a third when the monthly fee invoice is raised. Do not expect a one-to-one relationship between things that happen in the world and entries in a ledger. There isn’t one, and building software on the assumption that there is will produce a system that cannot represent a partial refund.

Correction three: balancing proves consistency, not truth#

This is the correction that matters most and that is stated least often.

A trial balance that agrees tells you exactly one thing: that for every debit posted, an equal credit was posted. It tells you nothing whatsoever about whether the entries were correct.

Post a £4,000 laptop purchase to “office refreshments” instead of “computer equipment”, and the trial balance agrees. Post an entire month’s rent to nowhere at all — leave it out of the books completely — and the trial balance agrees. Post a sale as £810 when the invoice said £108, on both sides, and the trial balance agrees. Swap the debit and the credit so a payment out looks like a payment in, and the trial balance agrees. Make two independent mistakes of £20 that happen to fall on opposite sides, and the trial balance agrees.

Every one of those has a name in British accountancy training, and the whole class of them — errors of principle, omission, commission, original entry, reversal and compensation — is invisible to the balancing check. The technical section tabulates them.

This has a direct implication for anyone building a payment system. A balanced ledger is a necessary condition of correctness and nowhere near a sufficient one. Balance is cheap: any halfway competent implementation enforces it at write time and it will never fail again. The expensive question is whether the entries describe what actually happened, and that question is answered not by the ledger but by comparing the ledger against an independent record produced by somebody else. That comparison is reconciliation, and it is where the real work is.

Fraud, it should be said plainly, generally balances beautifully. A dishonest employee does not create unbalanced entries. They create balanced entries to the wrong account.

Correction four: there is no world ledger, and the two halves are not simultaneous#

The plain version quietly implies a single book in which both ends of a journey get written. For a cake stall, that is true. For payments, it is not true and never has been.

There is no global ledger of money. There are millions of separate ledgers — one per bank, one per payment institution, one per scheme, one per merchant, one at the central bank — each of which balances within itself and each of which is entirely capable of disagreeing with its neighbours. When Ama’s bank debits her account and credits an amount payable into the card scheme, that is complete and balanced double entry inside Ama’s bank. It says nothing about whether the acquirer at the other end has recorded the matching amount receivable, or recorded it for the right value, or recorded it at all.

The two halves are also frequently separated in time, sometimes by days. Ama’s card was approved in about a second at 07:42 on a Tuesday. No posting was made at 07:42. Her bank placed a hold, which reduced her available balance but did not touch her ledger balance, and the actual double entry was made when the transaction cleared — typically the next business day for a UK card payment. That gap is not a defect. It is a designed consequence of authorisation and clearing being separate events, and it is the reason “pending” exists on your app.

Between the two halves, the money lives in accounts that exist purely to hold incomplete journeys: suspense accounts, clearing accounts, in-transit accounts. A large payment institution’s suspense accounts are where its operational health can be read most honestly, because anything sitting in suspense is, by definition, a journey with only one end written down so far. A suspense balance that is growing rather than clearing is an incident in progress.

The technical version#

The accounting equation#

Everything in double entry descends from one identity. In the form used in the United Kingdom and internationally:

Assets = Liabilities + Equity

An entity’s resources must equal the claims on those resources. The claims come from two sources: outsiders (liabilities) and owners (equity). This is not an empirical finding about businesses. It is true by construction, because equity is defined as the residual.

The IASB’s Conceptual Framework for Financial Reporting, revised in March 2018, gives the definitions that UK and international reporting rests on. An asset is “a present economic resource controlled by the entity as a result of past events”. A liability is “a present obligation of the entity to transfer an economic resource as a result of past events”. Equity is “the residual interest in the assets of the entity after deducting all its liabilities”. Income is “increases in assets, or decreases in liabilities, that result in increases in equity, other than those relating to contributions from holders of equity claims”, and expenses are the mirror.

Because income and expenses are themselves changes in equity, the equation expands to cover trading:

Assets + Expenses = Liabilities + Equity + Income

Left-hand-side accounts have natural debit balances. Right-hand-side accounts have natural credit balances. Every rule you will ever be asked to memorise about debits and credits is a restatement of which side of that equation an account sits on.

Debits and credits, defined properly#

A debit is an entry on the left of an account. A credit is an entry on the right. The words carry no directional meaning about cash. Their effect depends entirely on the account type:

British bookkeeping students learn this as DEAD CLIC: Debits increase Expenses, Assets and Drawings; Credits increase Liabilities, Income and Capital.

The etymology is worth knowing because it explains the counter-intuitive part. “Debit” comes through Old French from Latin debitum, the past participle of debere, “to owe” — recorded in English from the mid-fifteenth century, and in its bookkeeping sense of an entry in an account by 1776. “Credit” arrives in the 1540s from Italian credito, from Latin creditum, “a loan, a thing entrusted”, from credere, “to trust”. The Italian merchant’s ledger was originally a list of people: those who owed him (debitori, on the left) and those he owed or who had entrusted him (creditori, on the right). Every other account type was later fitted into that same two-column frame. That is why the words feel wrong when applied to a cash account: they were never about cash in the first place.

Books of prime entry, ledgers and posting#

The structure Pacioli described in 1494 is, with different names and vastly better tooling, the structure in use today.

A transaction is first captured in a book of prime entry — the day book, the sales day book, the cash book, or in a modern system the transaction log of the till, the gateway or the API. Nothing is classified yet; it is a chronological record that something happened.

It is then written as a journal entry: a dated, balanced set of debits and credits with a narrative and a reference. The journal entry is the atomic unit of accounting. It either exists in full or not at all.

The journal entry is posted to the general ledger, which is organised by account rather than by date. The list of available accounts is the chart of accounts, and its design is one of the more consequential decisions in a finance system: too coarse and you cannot answer questions, too fine and nobody classifies consistently.

High-volume accounts are broken out into subsidiary ledgers — one account per customer, per merchant, per cardholder — summarised into a single control account in the general ledger. The sub-ledger must sum to the control account. That check, run daily, is the first line of defence in every core banking system on earth, and its failure is a severity-one incident.

Finally, the trial balance lists every ledger account balance in two columns and proves the totals agree.

The trial balance and what it does not catch#

Where a difference is found and cannot be resolved immediately, the difference is parked in a suspense account so that reporting can proceed, and cleared as the errors are identified. A suspense account is an admission written into the books; a suspense balance that persists across period ends is a finding, not a feature.

Worked example one: £3.20 over the counter#

Kalypso Coffee, VAT-registered, sells a flat white for £3.20 including VAT at the UK standard rate of 20 per cent, paid in cash. Ingredient and packaging cost is 41 pence.

The VAT-exclusive value is £3.20 ÷ 1.20 = £2.6667, rounded to £2.67, leaving output tax of £0.53.

Journal 1 — the sale:

Journal 2 — the cost of what was sold:

Gross margin on the cup is £2.26. Note that the shop’s cash rose by £3.20 but its equity rose by only £2.26, because 53 pence was never its money and 41 pence of stock was consumed. A single-entry record showing “+£3.20” would overstate the shop’s performance by 42 per cent.

Worked example two: the same £3.20 across two banks#

Ama taps a contactless consumer debit card issued by Lloyds Bank. Kalypso Coffee’s acquirer is Worldpay; the merchant’s bank account is with NatWest; the card scheme is Visa. Ama’s bank and the acquirer’s settlement bank both hold accounts at the Bank of England.

For clarity the fee figures below are illustrative of a small UK merchant’s contract rather than quoted from a published rate card. The one fee that is not illustrative is the interchange cap: under the Interchange Fee Regulation as applied in the United Kingdom, interchange on a consumer debit card transaction is capped at 0.2 per cent of transaction value, and on a consumer credit card at 0.3 per cent, for transactions where the merchant, the acquirer and the issuer are all in the UK. Those caps have applied since 9 December 2015 and remain in force as of August 2026. Suppose Worldpay charges Kalypso a blended merchant service charge of 1.5 per cent.

Interchange at 0.2 per cent of £3.20 is £0.0064 — sixty-four hundredths of a penny. Merchant service charge at 1.5 per cent is £0.048. Both are computed to sub-penny precision and rounded once, at the level of the daily batch, not per transaction. This is not fussiness; it is the reason Volume I closes with a chapter insisting that money must never be represented in floating point.

Issuer (Lloyds), on clearing:

Lloyds’ liability to Ama falls by £3.20 — a debit, because reducing a liability is a debit, and this is the entry that prints the word “debit” on her statement. Lloyds retains interchange and owes the remainder into the scheme.

Acquirer (Worldpay), on clearing:

and then, for its own charge:

leaving £3.152 owed to the merchant. Note that interchange is revenue to the issuer and cost to the acquirer: the same 0.64 pence appears as a credit in one institution’s books and a debit in another’s. That is the four-party model in miniature, and Volume III takes it apart properly.

Merchant (Kalypso Coffee), at the till:

The shop has made a sale but holds no cash. It holds a receivable. This is the single most under-appreciated fact about card acceptance and the reason card-heavy businesses can be profitable and still fail: revenue is recognised at the till, cash arrives days later, and the gap must be funded.

Merchant, on funding the following business day:

Cardholder (Ama), if she keeps personal books:

Set Ama’s entry beside Lloyds’ entry and the whole chapter is visible in two rows. The identical event is a debit to the account in the bank’s ledger and a credit to the account in hers. Neither party is confused. They are keeping mirror books, and mirrors reverse.

Where the central bank’s books balance too#

The entries above leave one obligation outstanding: Lloyds owes money into the scheme, and the acquirer’s side is owed money out of it. That obligation is not discharged by the card network. Card networks calculate; they do not settle. Settlement happens when the participating banks move balances at a central bank.

In sterling, that means the Bank of England’s Real-Time Gross Settlement system. Institutions hold either a reserves account — available to banks, building societies and certain investment firms, remunerated at Bank Rate as set by the Monetary Policy Committee — or a settlement account, available to institutions that do not qualify for reserves but meet the Bank’s settlement criteria, including payment system operators and non-bank payment service providers.

The Bank supports several settlement models. CHAPS uses real-time gross settlement: each payment settles individually and irrevocably, debiting one participant’s account and crediting another’s, throughout the operating day. CHAPS is usually open from 06:00 to 18:00 Monday to Friday excluding England and Wales bank holidays; participants must be open to receive by 08:00 and to send by 10:00, customer payments must be submitted by 17:40 and the system closes to bank-to-bank payments at 18:00. There are over 35 direct participants, with several thousand further institutions reaching CHAPS indirectly. Retail systems, including card schemes, generally use deferred or prefunded net settlement instead, in which a day’s obligations are netted down to a handful of positions and settled in cycles — the subject of the netting chapter later in this volume.

Whichever model is used, the entry at the Bank of England has the same shape:

And here is the observation that ties the two-tier system together, which the chapter on central bank money will develop. Both of those accounts are liabilities of the Bank of England. When one commercial bank pays another, the Bank of England’s total liabilities do not change by a penny. Nothing is created and nothing is destroyed. The Bank simply records that it owes a different set of people. An interbank payment, at the apex of the system, is a relabelling exercise — which is precisely what the previous chapter said money was.

The gap between authorisation and posting#

Ama’s card was approved in roughly a second. Nothing was posted to her account in that second.

Authorisation is a request for, and a promise of, funds. In double-entry terms it is not a transaction at all: no debit, no credit, no change to the ledger balance. What the issuer records is a hold or shadow entry, which reduces the available balance — the number the app shows — while the ledger balance, the accounting figure, is untouched. Only at clearing, when the acquirer submits the transaction for settlement, does a balanced journal entry come into existence.

This produces effects that customers experience as inexplicable and that are in fact perfectly determinate. A pending transaction can disappear without ever posting, because a hold expired and no clearing record arrived. A refund appears slowly because it is a fresh transaction travelling the same route in reverse, requiring its own clearing and settlement, rather than a cancellation of the original. Two identical taps a second apart can both be approved because authorisation is checked against available balance, not against a queue of intentions.

Every one of these is the same underlying fact: authorisation and posting are separate events, and only one of them is double entry.

Mirror accounts, and why reconciliation exists at all#

Because each institution keeps its own books, the linkage between institutions is always a pair of accounts that are supposed to be equal and opposite. Lloyds’ “scheme settlement payable” should correspond to a scheme receivable; Worldpay’s “merchant payable — Kalypso” should correspond, to the penny, to Kalypso’s “card receivable — Worldpay”.

In correspondent banking the pattern has its own vocabulary. An account a bank holds with another bank, in that bank’s books, is a nostro — from the Italian for “ours”, as in “our account with you”. The same account viewed from the other side is a vostro — “yours”, as in “your account with us”. Banks additionally maintain an internal mirror nostro: their own running record of what the correspondent’s statement ought to say. The daily job of proving that the mirror matches the correspondent’s statement is nostro reconciliation, and Volume IV returns to it.

The general principle holds everywhere. Two balanced ledgers can be individually perfect and jointly wrong. Only comparison detects that, and comparison is a separate discipline with a separate cost. A payment institution that has invested in a beautiful ledger and not in reconciliation has built half a system.

The balancing invariant in software#

Several practical rules follow for anyone implementing this, and they are not stylistic preferences.

Enforce the invariant at write time. A ledger service should refuse to persist a journal entry whose debits do not equal its credits. The check belongs in the same transaction as the write, not in a nightly job. Once enforced, it never fails again, which is exactly the point: correctness that costs nothing to maintain.

Make postings immutable and append-only. Never update a posted entry. To correct an error, post a reversing entry and then post the correct one, leaving all three visible. This preserves the audit trail, makes any historical balance reproducible by replaying entries up to a point in time, and is the reason regulators can ask what a balance was on a given afternoon eighteen months ago and receive an answer.

Use signed integers in minor units. Amounts are held in pence, cents or paise as integers, with the currency and its exponent carried alongside per ISO 4217. Floating point must never touch a monetary amount. The final chapter of this volume is about nothing else.

Make posting idempotent. Every entry carries a client-supplied idempotency key. A retried API call after a timeout must not create a second, equally balanced, entirely spurious journal entry. Double-posting is the most expensive bug in payments precisely because the ledger will accept it without complaint: it balances.

Support multi-leg entries natively. A model that hard-codes exactly one debit account and one credit account cannot express VAT, fees, foreign exchange spreads or partial refunds without contortion. Purpose-built ledger databases — TigerBeetle is the best-known open-source example — take the opposite approach and expose only strict two-account transfers, composing complex movements from several of them. Either design is defensible; what is not defensible is a design in which balance is optional.

Model the incomplete journeys explicitly. Give in-flight money real accounts: authorisation holds, clearing accounts, unsettled scheme positions, unallocated receipts. Every penny in the business should be in a named account at all times, including the pennies whose ownership is currently undetermined. Systems that leave in-flight money implicit discover it during an incident, which is the worst possible moment.

What the law requires#

Double entry is not merely good practice in the United Kingdom; the outcome it produces is a statutory duty.

Section 386 of the Companies Act 2006 provides that “every company must keep adequate accounting records”, and defines adequate as sufficient “to show and explain the company’s transactions” and “to disclose with reasonable accuracy, at any time, the financial position of the company at that time”. Subsection (3) requires “entries from day to day of all sums of money received and expended by the company” together with “a record of the assets and liabilities of the company”. Companies dealing in goods must additionally keep year-end stock statements and stocktaking records.

Section 388 requires those records to be kept at the registered office or such other place as the directors think fit, open to inspection by the company’s officers, and preserved for three years from the date they were made in the case of a private company and six years in the case of a public company.

The statute does not mandate double entry by name. It does not need to. A record that shows and explains transactions and discloses the financial position at any time is a double-entry record; no other method produces it. For regulated firms the requirements go considerably further — the FCA’s safeguarding rules for payment and e-money institutions impose specific internal and external reconciliation duties on top of the Companies Act baseline — and Volume V takes those up.

Why this is the load-bearing chapter#

Every mechanism in the rest of this book is a special case of what is above.

Clearing is the exchange of the information needed to make matching entries in two institutions’ books. Settlement is the pair of entries at the central bank that discharges the resulting obligation. Netting is the arithmetic that reduces a million pairs of entries to one pair. Interchange is a credit in the issuer’s ledger and a debit in the acquirer’s. A chargeback is a reversing entry travelling the original route backwards. Reconciliation is the comparison of one institution’s balanced books against another’s. Safeguarding is the requirement that a specific pool of assets equal a specific pool of liabilities at all times. An outage is a period during which one side of a set of journeys was written and the other was not; the incident is the outage, and the aftermath is finding the unmatched halves.

There is no point in the entire apparatus at which anything is transported. There is only a very large number of books, each of which must balance internally, all of which must agree with each other, and none of which is allowed to record half a journey.

Pacioli’s advice was that a merchant should not sleep until the debits equal the credits. Modern payments run continuously and nobody sleeps at all, which is why the check has been moved into the machinery. But it is the same check, and it is asking the same question it asked in Venice in 1494: is there anything here that came from nowhere?

3.98 Common wrong ideas#

1. Wrong: debit means money going out and credit means money coming in. Right: debit is the left-hand side and credit is the right-hand side, so money arriving in a cash tin is a debit to it and money leaving is a credit.
2. Wrong: your bank statement is written from your point of view. Right: it is written from the bank’s, where your account is a liability, so the bank reducing what it owes you is a debit and that is the word it prints.
3. Wrong: “we will credit your account” is an unambiguous instruction. Right: it means nothing until you know whose book and what class of account, which is why reversed postings are the most tedious class of bug in financial software.
4. Wrong: double entry means every transaction has exactly two lines. Right: it means the sum of the debits equals the sum of the credits, and one entry may have three legs, or twelve, or two hundred.
5. Wrong: one event in the world produces one entry in the ledger. Right: a single coffee sale produces one entry for the money and another for the stock, with further entries at settlement and at invoicing, and software built on a one-to-one assumption cannot represent a partial refund.
6. Wrong: a trial balance that agrees means the books are correct. Right: it means only that for every debit posted an equal credit was posted, and a whole month’s rent left out entirely leaves the totals untouched.
7. Wrong: fraud shows up as an imbalance. Right: a dishonest employee does not create unbalanced entries but balanced entries to the wrong account, which is why reconciliation exists as a discipline separate from bookkeeping.
8. Wrong: there is a single global ledger into which both halves of a payment are written. Right: there are millions of separate ledgers, each balancing within itself and each entirely capable of disagreeing with its neighbours.
9. Wrong: the two halves of a payment happen at the same moment. Right: the card was approved in about a second and the double entry was made at clearing, typically the next business day, with the money sitting in suspense and clearing accounts in between.
10. Wrong: a ledger that enforces balance is a correct ledger. Right: balance is cheap and never fails again once it is enforced at write time; the expensive question is whether the entries describe what actually happened, and only comparison against somebody else’s record answers it.

3.99 Chapter summary in 20 lines#

1. “I spent three pounds twenty on a coffee” sounds like one fact and is at least two, and for the shop it is four or five.
2. The rule of double entry is that you never write down one end of a journey, only both ends together.
3. Debit means the left-hand column and credit means the right-hand column, and the words mean nothing more than that.
4. Because the two column totals must agree, the method checks itself, which let a shopkeeper in 1350 catch his own errors before they compounded.
5. Aisha’s cake stall reaches £34 by two entirely separate routes, and that is a consequence of the rule rather than luck.
6. Almost everyone in Britain believes debit means money out, because bank statements are written from the bank’s point of view and never explain whose books they are.
7. In the bank’s ledger your account is a liability, so spending reduces it, and reducing a liability is a debit.
8. In your own books that account is an asset, so the identical event is a credit: the two parties keep mirror books, and mirrors reverse.
9. “Double” refers to the two sides rather than to two lines, and the real rule is that the sum of the debits equals the sum of the credits.
10. A VAT-registered shop’s £3.20 flat white needs three legs, because 53 pence of it was never the shop’s money.
11. Some real-world events need several separate journal entries, so there is no one-to-one relationship between happenings and entries.
12. A trial balance that agrees proves internal consistency and says nothing whatever about whether the entries are true.
13. Errors of omission, commission, principle, original entry, reversal and compensation are all invisible to the balancing check.
14. Fraud generally balances beautifully, because the dishonest entry is balanced and merely posted to the wrong account.
15. There is no world ledger: millions of separate books each balance internally and can still disagree with one another.
16. Following one £3.20 card payment through issuer, acquirer, merchant and cardholder shows the same interchange as a credit in one institution’s books and a debit in another’s.
17. The merchant records a receivable rather than cash, which is why card-heavy businesses can be profitable and still fail on the funding gap.
18. Authorisation posts nothing and touches only the available balance, so pending items can vanish, refunds travel the whole route again, and two identical taps can both be approved.
19. At the apex, an interbank payment moves balances between two accounts that are both liabilities of the Bank of England, so nothing is created and nothing is destroyed.
20. Clearing, settlement, netting, interchange, chargebacks, reconciliation and safeguarding are all special cases of the same refusal to record half a journey.

Chapter sources: Luca Pacioli, Summa de arithmetica, geometria, proportioni et proportionalità (Venice: Paganino de Paganini, November 1494), tractatus XI of distinctio IX, “Particularis de computis et scripturis”; Mathematical Association of America, Convergence, “Mathematical Treasure: Luca Pacioli’s Summa”; Institute of Chartered Accountants in England and Wales, historical accounting literature collection, entries on Pacioli and the Summa; G. A. Lee, “The Coming of Age of Double Entry: The Giovanni Farolfi Ledger of 1299–1300”, Accounting Historians Journal 4(2), 1977; Johann Wolfgang von Goethe, Wilhelm Meister’s Apprenticeship, Book I, Carlyle translation. IFRS Foundation, Conceptual Framework for Financial Reporting, revised March 2018, project summary. Companies Act 2006, sections 386 and 388 (legislation.gov.uk). Bank of England, “A brief introduction to the Real-Time Gross Settlement system and CHAPS”, and the CHAPS operating-hours pages. Payment Systems Regulator, guidance on the Interchange Fee Regulation. Association of Accounting Technicians, study guidance on identifying and correcting errors. Online Etymology Dictionary, entries for “debit” and “credit”.