Two things in this dataset should not sit next to each other. The first is the level of precision. The second is the level of explanation.

The most recent figures available from the Ministry of Justice's prison custody deaths publication give a single value for 2024: 18.0. Not 18. Eighteen point zero. Precision to one tenth (Source: Ministry of Justice, Safety in Custody -- Deaths_in_prison_custody_1978_to_2024_accessible -- Table_1_10).

Now the second thing. The heading this figure sits under is the word "Window". That is the label. That is all of it. A number recorded to a tenth of a unit, filed beneath a word that tells a reader nothing about what is being counted, or of what, or per what.

Hold those together for a second, because the tension is the story. Decimal places are a claim about confidence. They say: we know this closely enough that the last digit matters. A column heading is a claim about meaning. It says: here is what the number is. One of these has been done to a high standard. The other has not been done at all.

Here is the strongest honest alternative reading, and it matters. If 18.0 is not a count of individuals but a rate, a percentage or an average, then the decimal is entirely correct and my complaint is misplaced. And "Window" may well be a truncated or mangled header pulled from a wide spreadsheet, rather than the phrase the Ministry of Justice intended anyone to read. Machine extraction from complex statistical tables goes wrong in exactly this way. On the evidence available here, we cannot say which of those explanations applies, and that is the point: the figure does not carry enough with it to be read safely.

Which brings us to what this data cannot tell you. It cannot tell you whether prisons in England and Wales got safer in 2024. It cannot tell you whether the figure went up or down, because the extract available gives one value for one year and nothing to compare it against. If the number improved, that would be genuinely good news and it deserves saying out loud. We are not in a position to say it.

That is the frustrating part. Safety in custody statistics are among the most consequential numbers the state publishes about itself. They concern people the government has taken into its own care and is legally responsible for keeping alive. Numbers like that get quoted in inquests, in inspection reports, in coroners' findings. They should survive being lifted out of a spreadsheet without losing their meaning.

My view, and I'll own it: a statistic about deaths in state custody that arrives labelled "Window" has failed before anyone has even argued about what it shows. Accuracy to one decimal place is worthless if the reader has to guess the unit.

One last note on timing. 2024 is the most recent year available in this publication, so this is a look backwards, not a snapshot of prisons today. Whether the following year moved in the right direction, we cannot tell from what has been published here.