Skip to content
Sattari

The Horizon Scandal — Branch Accounting and Enforcement at the UK Post Office

A structural reading of the Horizon settlement framework

The United Kingdom’s Post Office does not staff most of its branches. It contracts other people to run them, and it built a system to account for the money each branch holds and to settle any difference. Horizon, the software at the center of the scandal, was one part of that system: the part that produced the figure a branch was measured against. This reading sets the software aside at the start and asks the prior question the method asks of anything built: what was the whole system meant to do, and what must any system built to do that contain? Worked forward from the system’s own purpose, the failure is not the software errors the courts later found. It is a function the design never included, absent on the day the system was drawn, years before the first error made it visible.

The system that was built

The United Kingdom’s Post Office runs a network of some thousands of branches. It does not own or staff most of them. Instead it contracts individual operators, each called a subpostmaster, to run a branch: usually a local shop whose owner also provides Post Office counter services. Each branch handles cash and holds Post Office property, such as stamps, postal orders and foreign currency. The operator is trusted with all of it, and the money belongs to the Post Office, not to the operator.

This creates a problem the Post Office has to solve. It is responsible for money that is handled every day by thousands of people it does not employ and cannot watch. If a branch comes up short, the Post Office needs a way to find out, and a way to be made whole. So it built a system to do exactly that, and that built system is what this reading examines.

The system works like this. Every branch runs on an electronic accounting system called Horizon, supplied to the Post Office by Fujitsu, a technology contractor. Horizon records the branch’s transactions and keeps a running figure of the cash and property the branch ought to hold. At regular intervals the operator counts what is actually at the branch and compares it with Horizon’s figure. If the two match, the branch balances and carries on. If they do not, Horizon shows a shortfall, and here the system does the thing it was built to do: it requires the operator to settle that shortfall, either by paying the difference at once or by accepting it as a debt owed to the Post Office. The operator’s contract makes the operator personally liable for it. Where a shortfall is not settled, the Post Office may investigate it, recover the money as a debt, or pursue other legal remedies, including prosecution.

Read as a built thing, this is a system for settling each branch’s account — for deciding what the branch truly holds, and resolving any difference between that and what Horizon says it should hold. A branch’s money and property move every day, through the hands of an operator the Post Office cannot directly supervise and across transactions that originate well beyond the counter, so the Post Office needs a way to strike each branch’s true position and to recover a real loss when there is one. For most branches most of the time, the system did exactly that.

The system under reading is the one just described. The line is drawn around the work, not around the company. Fujitsu is a separate firm, and until 2012 the enforcement ran through Royal Mail Group, which the Post Office was then part of. Both are inside the line. The figure the whole system turns on came from Fujitsu’s software, and the cases were brought in Royal Mail Group’s name.

Three things sit just outside. The first is the courts. The system handed them its cases, but what a court made of a case is not something the Post Office built, so the reading stops at the courthouse door. The second is a rule of law: a court would treat a computer as working correctly unless someone showed otherwise. The system leaned on that rule but did not make it. The third is the government. It owns the Post Office but did not build or run this system.

What settling an account requires

Begin where the method begins, with the purpose, and ask what a system must contain to serve it. To settle a branch’s account is to decide what the branch truly holds and to resolve any difference from what it should hold. Four functions follow from that purpose. They can be named before looking at anything the Post Office actually built, because they follow from the purpose itself.

First, a way to work out what the branch should hold: a figure, built from every transaction that passed through the branch.

Second, a way to establish what the branch does hold: a count of the cash and property actually there.

Third, a way to decide which is right when the two disagree. A gap between the figure and the count is not yet a loss. It is a question. The money may truly be missing, or the figure may be wrong. Only by answering that question can the system settle the branch to its true position, because until it is answered there is no true position, only two numbers that do not match.

Fourth, a way to act on the answer: recover a real loss, or correct a false figure.

The third function is the one the purpose turns on. It is what makes the word “true” in the purpose mean anything. A system can measure both sides, and can act, and still not have it. Remove it, and the system can no longer settle to the truth. It can only settle to whichever of the two numbers it was built to believe.

The function the system did not have

Three of the four functions were built, and built well. Horizon produced the figure. The operator’s count established what was present. And the means of acting on a difference were extensive. They included: paying at once or settling centrally over time, the Post Office’s own corrections to the account, recovery of an unpaid shortfall as a debt, and, through much of the period, prosecution.

The third function was not built. The system had no component whose task was to decide, when the figure and the count disagreed, which was right. In its place stood a rule: the figure is right, and the difference is the operator’s to settle.

At the end of each accounting period the operator had to resolve any shortfall before the branch could carry on. The High Court set out the only two routes the system offered: the contract “require[s] the SPM to ‘make good’ the shortfall, which means pay the money to the Post Office,” or the operator could “settle centrally,” which “means seek time to pay.”1 Both are payment. The court found that such a shortfall “would be treated as a debt by the Post Office,”1 and that the operator’s “choice, such as it was, was either paying immediately or settling centrally,” with “no real ‘choice’” at all.1 Nowhere between the figure and the payment did the system stop to ask whether the figure was right.

That the figure could be wrong was not a remote possibility the design might reasonably ignore. The account was built partly from entries the operator did not make and could not refuse. The court found that transaction acknowledgements arriving from outside the branch “simply have to be accepted by the SPM, and could not be challenged.”1 So the operator was made liable, at the period’s end, for a figure that other hands had helped write and that the operator had no means to test.

In Sattari’s reading of the case, this is the first and most upstream finding. It is not that the figure was sometimes wrong. It is that the system was built to settle against the figure whether or not it was right, because it had nothing that could tell the difference. A system built this way behaves the same whether its figure is sound or not. The only difference is that, while the figure happens to be right, no one is harmed. The fault is therefore present on the day the system is drawn, complete, before a single error occurs. It is a property of the arrangement, not of any part within it, which is why it is located at the whole. What follows can be seen as aspects of this absence, seen at a different place in the system.

What follows from the absence

A missing function does not leave an empty space. The work it would have done still has to be done by something, and in this system it was done by default, in a particular shape, everywhere a difference arose.

If the third function had existed, it would have needed evidence to work with, and that evidence existed. Horizon kept a complete record of what had actually happened at the branch, down to each keystroke. The High Court found this record to be “the best evidence available of what has occurred on Horizon,” and held that it “should be consulted in circumstances where there is a dispute between the Post Office and a SPM.”1 It was not. The experts agreed, and the court found, that “the Post Office does not consult the full audit data … before deciding how to handle discrepancies and issuing Transaction Corrections.”1 The one record that could have established the truth of a figure was not looked at before the figure was acted on.

That was not an oversight so much as a consequence of where everything sat. The record that could test the figure was held by the firm that produced the figure and by the party the difference was owed to. The knowledge of whether the figure-maker had erred lived in the same two places. The operator, the one party with an interest in asking whether the figure was right, had none of the means to ask, and the parties with the means had the opposite interest. The Post Office was at once the author of the figure, the investigator of the shortfall, the party owed the money, and, until about 2015, the prosecutor.2

And because each operator answered only for their own branch, no one stood in a position to see the one pattern that would have exposed a figure-side cause: the same discrepancy surfacing in many branches at once. A fault in the shared system shows itself across the network. A system divided branch by branch, with each operator liable alone, has no reader for that signal. The division made each branch visible and the network invisible.

A system that settled to the truth would learn from what it found. A figure shown to be wrong would correct the figure, and would raise a question about every other figure produced the same way. This system had no such path, because the finding that would feed it, that a figure was wrong, was the finding it had no function to reach.

In the place of that correction, a different loop ran, and it ran the wrong way. Each shortfall paid and each prosecution won was taken as confirmation that the figure had been sound. The Court of Appeal described the effect: “As each prosecution proceeded to its successful conclusion the asserted reliability of Horizon was, on the face of it, reinforced,” and “the longer that approach persisted, the more POL was able to, and did, rely upon its own past abusive conduct by asserting that no previous challenge to Horizon had succeeded.”2 No challenge had succeeded because none could, and that was read as proof that none should. A system with no way to discover it was wrong had, in its place, a mechanism that made it more sure of itself the longer the error ran.

When the fault became live

A system that settles against its figure without testing it is harmless for exactly as long as the figure is right. The fault described so far is latent — built in, but dormant while Horizon’s numbers happen to match reality. What woke it was that they did not.

The High Court found that bugs, errors and defects in Horizon could “cause apparent or alleged discrepancies or shortfalls relating to Subpostmasters’ branch accounts,” and that this was not merely possible: “it actually has happened, and on numerous occasions,” across both the older and the newer versions of the system.1 It found “a significant and material risk on occasion of branch accounts being affected” this way.1 It found the original Horizon “not remotely robust.”1 It found that the system “did not alert SPMs” when this happened.1 And it found that Fujitsu could, and in one respect the Post Office could, “insert, inject, edit or delete transaction data or data in branch accounts … without the knowledge or consent of the SPM,” and that such an entry “would look as though the SPM had done it.”1

These are the facts the public account of the scandal is built on, and they are true. But they are not the fault; they are the occasion on which the fault produced harm. A system with the third function would have caught every one of these errors, because each produced a figure that reality contradicted, and establishing that contradiction is precisely what the third function is for. The bugs mattered so much, and for so long, only because the function that would have caught them did not exist. Read as a system, the software error is downstream. Fix every bug and the design is unchanged. The next wrong figure, from any cause, lands the same way, on the operator, unexamined.

What it cost

Because the fault sat at the whole, it did not fail in scattered places. It failed everywhere the system ran, in the same direction every time, against the operator. The statutory inquiry found it likely that “about 1,000” people were prosecuted and convicted across the United Kingdom on Horizon evidence,3 and that thousands more were suspended over apparent shortfalls.3 In “very many” cases, it found, “liability was imposed upon postmasters for losses which simply did not exist.”3 The Post Office told the inquiry of thirteen people who, according to their families, took their own lives; the inquiry’s chair could not make a definitive finding of a causal connection and did not rule one out.3

The courts and Parliament have since undone the convictions. The Court of Appeal, quashing the first of them, found that “there was no evidence of a shortfall other than the Horizon data,” so that “if the Horizon data was not reliable, there was no basis for the prosecution.”2 In 2024 Parliament quashed the qualifying convictions by statute, a step its Explanatory Notes call “an unprecedented and wholly exceptional legal solution to a miscarriage of justice of unparalleled scale and impact.”6 By September 2026 the Ministry of Justice had identified 620 people in England and Wales with a conviction quashed by that Act, 2,250 convictions in all,7 and about £1,731 million had been paid in redress to more than 14,000 claimants.8 This is the measure of what a single missing function cost, compounded across the years the figure was wrong and the system had no way to say so.

What this reading does not claim

The record also establishes that the Post Office knew of Horizon’s problems and kept them from the people it was holding liable. The Court of Appeal found that “POL knew that there were serious issues about the reliability of Horizon” and “deliberately chose not to comply with its obligations” as a prosecutor.2 The High Court found that the Post Office should have told operators when a bug had hit their accounts, and that correcting an account “in the ordinary course” had meant “keeping the cause or reason for the correction secret.”1 It described the Post Office’s stance on Horizon as “a simple institutional obstinacy,” amounting to “the 21st century equivalent of maintaining that the earth is flat.”1 The inquiry found that the Post Office “maintained the fiction that its data was always accurate.”3 Its remaining volumes will address who knew what, and when.4

This reading makes no finding about any of that, because it is not what the method reads. The method reads the arrangement, not the conduct of the people in it. It does not detect deception and it does not assess blame. What it can say is narrower, and it is the reading’s own judgment: the concealment compounded the harm, but it did not create it. A system with no function to establish whether its figure was true would have held honest operators liable for losses that did not exist even if everyone working in it had acted in good faith, because the fault is the missing function, not anyone’s dishonesty. The concealment is, if anything, downstream of the same design. Once the whole system rested on the figure being beyond question, doubt about the figure threatened everything built on it. The Post Office’s own people saw this: an internal report in 2010, recited by the Court of Appeal, warned that “any perception that POL doubts its own systems would mean that all criminal prosecutions would have to be stayed.”2 A system that cannot test its figure cannot afford to let anyone else test it either.

The statutory inquiry has published only the first of its volumes. Its findings on the Post Office’s investigation and prosecution functions, and on who knew what and when, are still to come, and may add to the account of the parts this reading places at findings 2 and 3. They are unlikely to disturb the reading’s structure, which rests on the design of the system rather than on anyone’s conduct.

Where the reading cites what an operator’s contract said about the investigation function, it relies on an independent review’s quotation of that contract, not on the contract itself.

The reading concerns the system as it operated between about 1999 and 2015. It does not assess how the Post Office accounts for branch shortfalls today.

The direction of the remedy

Each finding points to a different kind of fix, and they are not interchangeable. The design fault calls for the missing function to be built: a way to establish, when the figure and the count disagree, which is right, before anyone is made to answer for the difference. The division calls for that function to be placed where neither the maker of the figure nor the party owed the money controls it, for the evidence that tests a figure to be given to the person held liable for it, and for someone to read the branches together. The missing loop calls for a wrong figure to feed back and correct the system, and for the system to stop treating the success of its own enforcement as proof that it was right. The difference between the figure and the count would then be held open, as a question, until the question was answered, rather than closed at once onto the operator.

Some of this has since been done from outside the system. From about 2015 the Crown Prosecution Service, not the Post Office, prosecuted operators.2 The Post Office says the old assumption that operators were wholly responsible for unexplained losses “was wrong,” and that its presumption today is the reverse.9 The government is reviewing the rule of law that let a court presume a computer correct.11 And it plans to replace Horizon altogether.10

Replacing Horizon answers the software errors. It does not answer the fault. A new system that produces the figure, placed in the same arrangement, would be read the same way — its figure taken as true, its difference placed on the operator, and with nothing built to test it. Whether the fault has been fixed is a question about the arrangement, not about the software, and only a reading of the whole can answer it.

The question that could have been asked at the start

Nothing in this reading needed the event to have happened. It needed only the design, and the question the design invites. That question could have been asked on the day the system was drawn, from the blueprint alone, before a single branch went live:

This system settles each branch by requiring the operator to answer for the difference between the count and the figure Horizon reports. What in it establishes which of the two is right when they disagree? And if nothing does, on what basis is the difference the operator’s, rather than the system’s own?

The answer, nothing, and no basis, was available from the start. It took no bug to find it, no conviction, and no court. It took only reading the system as a whole, against the purpose it was built to serve.

  1. 1

    Bates & Others v Post Office Ltd (No. 6: Horizon Issues) [2019] EWHC 3408 (QB)

    High Court of Justice, Queen’s Bench Division (Fraser J) · December 16, 2019 · Court judgment

    The system judgment, read in full from the approved text. Paragraphs relied on: 95 (transaction acknowledgements arriving from outside the branch ‘simply have to be accepted by the SPM, and could not be challenged’); 442 (the Post Office should have told operators of a bug; ‘corrected in the ordinary course’ meant keeping the cause secret); 804 (an operator’s ‘choice’ was to pay at once or seek time to pay; no real choice); 824 (a shortfall requires the operator to ‘make good’ or ‘settle centrally’); 825 (a shortfall ‘would be treated as a debt’); 905 (the Post Office does not consult the full audit data before deciding how to handle discrepancies); 911 (the audit data is ‘the best evidence available’ and ‘should be consulted’ where there is a dispute); 928-929 (‘institutional obstinacy’; the ‘earth is flat’); 968-970 (bugs could and did cause shortfalls, on numerous occasions; a significant and material risk); 973 (‘the Horizon system did not alert SPMs’); 975-978 (Legacy Horizon ‘not remotely robust’; a material risk of Horizon-caused shortfalls, 2000-2010 and 2010-2017); 1001-1005 (Fujitsu, and in one respect the Post Office, could alter branch data without the operator’s knowledge or consent; an injected transaction ‘would look as though the SPM had done it’).

    Retrieved November 6, 2026

  2. 2

    Hamilton & Others v Post Office Ltd [2021] EWCA Crim 577

    Court of Appeal, Criminal Division · April 23, 2021 · Appellate judgment

    The criminal appeals of 42 former operators and staff. Paragraphs relied on: 4 (each case pursued by the Post Office as a private prosecutor; convicted 2003-2013 for conduct in 2000-2012); 24 (the 2010 report warning that doubt in Horizon would mean prosecutions had to be stayed); 28 (the Crown Prosecution Service took over from about 2015); 121, 123, 129, 130, 137 (the Post Office knew of Horizon’s problems; no evidence of a shortfall other than Horizon’s; the prosecutions an affront to the conscience of the court).

    Retrieved November 6, 2026

  3. 3

    Post Office Horizon IT Inquiry, Final Report, Volume 1 (HC 1119)

    Post Office Horizon IT Inquiry (Sir Wyn Williams) · July 8, 2025 · Inquiry report

    The inquiry’s first volume, on human impact and redress. Paragraphs relied on: 1.4 (the Post Office maintained the fiction that its data was always accurate); 3.8 to 3.12 (thirteen people who, their families say, took their own lives; no definitive finding of causation, not ruled out); 3.24 (about 1,000 prosecuted and convicted across the UK); 3.26 (thousands suspended); 3.27 (in very many cases liability imposed for losses that did not exist).

    Retrieved November 6, 2026

  4. 4

    Progress update from the Chair

    Post Office Horizon IT Inquiry · July 8, 2026 · Inquiry statement

    The remaining five volumes of the final report are still in preparation.

    Retrieved November 6, 2026

  5. 5

    Interim Report into alleged problems with the Horizon system

    Second Sight Support Services Ltd, for Post Office Ltd · July 8, 2013 · Independent review

    The first independent review, commissioned by the Post Office. Section 7.4 quotes the operators’ contract: ‘The Investigation Division does NOT enquire into matters where crime is not suspected.’ Section 8.2(e): the lack of an effective outreach investigations function meant the Post Office failed to find the root cause of problems.

    Retrieved November 6, 2026

  6. 6

    Post Office (Horizon System) Offences Act 2024, with Explanatory Notes

    UK Parliament (legislation.gov.uk) · May 24, 2024 · Statute

    Quashed qualifying convictions in England and Wales and Northern Ireland on the day it came into force. The Explanatory Notes, prepared by the government and not endorsed by Parliament, call it (para. 23) ‘an unprecedented and wholly exceptional legal solution to a miscarriage of justice of unparalleled scale and impact.’

    Retrieved November 6, 2026

  7. 7

    Quashed convictions management information: 24 September 2026

    Ministry of Justice · September 24, 2026 · Official statistics

    England and Wales only; data as of 15 September 2026; the final release in the series. 620 people with at least one conviction quashed by the Act; 2,250 convictions in all.

    Retrieved November 6, 2026

  8. 8

    Post Office Horizon financial redress data as of 30 September 2026

    Department for Business, Innovation, Science and Trade · October 16, 2026 · Official statistics

    About £1,731 million paid across the schemes to more than 14,000 claimants.

    Retrieved November 6, 2026

  9. 9

    Horizon scandal: frequently asked questions

    Post Office Ltd · October 21, 2026 · Party statement

    The Post Office’s own statements: Horizon is still in use; the former assumption that operators were wholly responsible for unexplained losses ‘was wrong,’ and today the presumption is that they are not. A party’s statement, not a finding.

    Retrieved November 6, 2026

  10. 10

    Government response to the Green Paper: the future of the Post Office

    Department for Business and Trade · February 25, 2026 · Government response

    The government will ‘transition operation of Horizon away from Fujitsu and ultimately replace the Horizon system altogether.’

    Retrieved November 6, 2026

  11. 11

    Use of evidence generated by software in criminal proceedings: call for evidence

    Ministry of Justice · January 21, 2025 · Government consultation

    Describes the common-law presumption that a computer was working correctly unless shown otherwise, replacing a repealed statutory rule from 2000, and says its limitations were ‘highlighted starkly by the Post Office Horizon scandal.’ Opened 21 January 2025; closed 15 April 2025.

    Retrieved November 6, 2026

All publications