When the system is wrong: the evidence lesson from Horizon

Public record

The documented case

On 23 April 2021, the Court of Appeal considered 42 appeals in Hamilton & Others v Post Office Ltd. The official court summary records that 39 appeals succeeded and the corresponding convictions were quashed. Three appeals were dismissed: the court distinguished cases in which Horizon reliability was essential from those in which it was not. [1]

For the Horizon cases, the court identified significant system problems and failures of investigation and disclosure. Apparent accounting shortfalls could not simply be equated with missing cash or stock. That distinction sits at the centre of this historical decision. [1][2]

365 ALTERAM analysis

The forensic question: what does the record actually prove?

Our methodological lesson is narrower than a claim that computer evidence is generally unreliable. A number on a screen is an output. Before drawing a conclusion about conduct, an investigator needs to understand the inputs, processing rules, interventions and known limitations that produced it. The evidence must support the proposed inference, not merely resemble it.

A useful starting point is a proposition register. Write down the allegation precisely: a loss occurred; a particular user caused it; the act was deliberate. Then identify the material supporting each proposition and what would independently test it. One record may be relevant to several propositions without proving all of them.

365 ALTERAM analysis

A practical independent-review workstream

First, preserve the material received and record its origin, extraction date, format and handling history. A spreadsheet export should not silently stand in for the underlying system record. Request the data dictionary, relevant configuration, access history and available error or incident records through authorised channels.

Second, reconstruct a small number of material events end to end. Compare the source transaction with the application output, later adjustments and independently available banking or stock records. Record missing information explicitly. Where technical access is unavailable, state what cannot be tested rather than treating the absence of a discovered defect as proof of reliability.

Third, test competing explanations consistently. A processing defect, authorised adjustment, user mistake and deliberate act are different hypotheses. Each needs a statement of supporting material, contrary material, limitations and the next proportionate test. This prevents an attractive explanation from becoming an unexamined conclusion.

365 ALTERAM analysis

What counsel should receive

A useful report connects every important conclusion to its source and identifies the assumptions on which it depends. It should include an event chronology, an explanation of the data pathway, a discrepancy schedule and a list of questions still requiring disclosure or technical expertise. Findings that support the allegation belong in the report alongside findings that weaken it.

Hamilton does not determine the outcome of another case. Its value here is as a disciplined reminder: the reliability of a system and the meaning of a record require examination. Qualified counsel decides the legal implications; appropriately qualified specialists address technical questions within their competence.