Bates v Post Office

The brief

For years the Post Office maintained that its Horizon accounting system was robust, and that discrepancies in branch accounts were almost always the fault of the sub-postmaster rather than the technology. On that premise hundreds of sub-postmasters were pursued for shortfalls, many lost their livelihoods, and a large number were prosecuted for theft, fraud or false accounting.

The group litigation had to settle the technical question underneath all of it: could Horizon itself introduce or produce the very shortfalls that sub-postmasters were being held liable for, and was it as reliable as the Post Office had always claimed? Almost everything else depended on the answer.

A warning in 2003, years before the group action

I first looked at Horizon more than fifteen years before the group litigation reached trial. Around the end of 2003 I was approached by the firm Weightman Vizards and instructed, on a joint basis, in Post Office Ltd v Mrs J Wolstenholme, a county court claim concerning the Cleveleys sub-post office in Lancashire. By then I had spent years investigating failed and disputed technology projects as an accredited expert witness.

I was given a folder of roughly 90 printouts from the Horizon helpdesk and asked, in effect, whether the sub-postmistress had simply been using the system badly. I built a spreadsheet, worked through every call, and reached the opposite conclusion. Of the calls I analysed, I judged 63 to be, in the words of my report, “without doubt system related failures,” and only around a dozen to be the sub-postmaster contacting the wrong helpline. Most were screen locks, freezes and blue-screen errors: the signature of faulty hardware, software or interfaces, not of user error.

My opinion, set out on 21 January 2004, was unambiguous:

From a computer system installation perspective it is my opinion that the technology installed at the Cleveleys sub-post office was clearly defective in elements of its hardware, software or interfaces. The majority of the errors as noted in the fault logs could not be attributed to being of Mrs Wolstenholme’s making.

When Fujitsu responded and I was asked whether that answer changed my view, it did not. On 27 February 2004 I replied that my opinion “remains as stated in my original note,” and asked to see the underlying material: the raw data, the crash dumps, the known-error records. I was told, variously, that the data had been deleted, was unavailable, or would be destroyed. None of it was produced. The matter went no further at the time, and the warning sat unheard for the better part of two decades.

The scale of the litigation

Bates v Post Office was brought as a Group Litigation Order on behalf of 555 claimant sub-postmasters, and was heard in the High Court across a series of trials. The Common Issues trial examined the contractual relationship between the Post Office and its sub-postmasters; the Horizon Issues trial examined the reliability of the system itself. In 2016, then at IT Group, I was instructed by James Hartley of Freeths to act as the claimants’ independent IT expert on the Horizon Issues. It is that trial, and the technical evidence at its centre, that this account concerns.

Into the primary technical record

This time the door that had been closed in 2004 was open, and my analysis went to the primary technical record rather than to assertion. That meant Horizon’s source code, the PEAK incident records used by Fujitsu’s development teams, and the Known Error Logs (KELs) that catalogue faults known to the people who built and maintained the system.

My task was disciplined and specific: to establish, from that record, whether bugs, errors and defects existed, whether they were capable of reaching a branch’s accounts, and how often that could happen. The same evidence the sub-postmasters had never been shown was now the foundation of the case against the Post Office’s account of its own system.

What the evidence showed

The evidence identified a series of bugs, errors and defects in Horizon capable of producing discrepancies in branch accounts, among them well-documented defects such as the Dalmellington and Callendar Square (Falkirk) bugs, and problems in the way the system handled suspense accounts and the reconciliation of receipts and payments. Individually and cumulatively, they contradicted the Post Office’s central claim that the system was robust and its figures reliable.

Equally important was the question of remote access: whether the Post Office or Fujitsu could insert, alter or delete transactions in a branch’s accounts without the sub-postmaster’s knowledge or consent. The technical record showed that such access existed. That removed a second pillar of the Post Office’s position: the assumption that a sub-postmaster was necessarily responsible for whatever their branch account showed.

I was cross-examined over four days on this evidence, and it held.

The judgment

In his December 2019 judgment, Mr Justice Fraser found that Horizon was not remotely as reliable as the Post Office had maintained, that it contained bugs, errors and defects, and that these could, and sometimes did, cause discrepancies in branch accounts. He preferred the evidence I had given for the claimants, and was openly critical of the Post Office’s expert for approaching the role as an advocate for his client rather than as an independent analyst assisting the court.

The judge was also critical of the Post Office’s conduct and its approach to disclosure, to the point that he referred matters arising from the case to the Director of Public Prosecutions. The Horizon Issues judgment underpinned the settlement of the group litigation and, in the period that followed, the quashing of scores of criminal convictions and the establishment of a statutory public inquiry into the scandal.

Why it matters

The Horizon Issues judgment is now a reference point for how rigorous technical evidence decides high-stakes disputes. It established, on the record and on the facts, that a computer system relied upon by a large institution could produce the very errors that institution had spent years denying, and that independent, source-level analysis was what made the difference.

It is also a lesson in what independence costs and why it counts. The conclusion the court accepted in 2019 was, in its essentials, the same one I had offered and seen set aside in 2004. An expert’s duty is to the court, and evidence has to be built to survive scrutiny, not to please the party instructing it. That principle is at the centre of how we work at EVO.