Reading a UK postcode crime report.

A postcode crime report condenses 12 months of police-recorded crime in a half-mile radius around a UK postcode into something a non-expert can read in two minutes. The figures look simple — a headline total, a list of categories, a small chart, a paragraph of summary — but every one of them carries assumptions worth understanding before any conclusion is drawn from them. This piece walks through every section of the report in order. For each, we explain what the number means, how it is produced, and, where it matters most, what it does not mean. By the end, the reader should be able to open a report and read it the way the data warrants: carefully, and without overreading any single figure.

Where the data comes from

Every count in the report originates from the Police UK open data API, the official public release of recorded crime in England and Wales. Forty-three territorial police forces contribute incident data on a monthly cadence. The API publishes four things for each incident: a category, the month in which it was recorded, an outcome status if one is available, and a street-level location that has been snapped to an anonymous map point on or near a public road. The map point is deliberately approximate — Police UK obfuscates exact coordinates so that no individual address can be identified from the open data.

Crime Insights queries this API for every crime within a half-mile radius of the postcode centroid and aggregates the results. We do not enrich the records with private data, we do not infer identities, and we do not retain anything beyond the public release. If a count surprises the reader, the same number can be reproduced by querying the Police UK API directly for the same area and month range.

The headline figure: total crimes in 12 months

The number at the top of the report is the sum of every reported crime in the radius across the most recent 12 months for which data is available. It is the easiest figure to read and the easiest one to misread, so it is worth being precise about what it represents.

It is a count of incidents that were reported to a police force, recorded by that force, and released through the open data feed. It is not a measure of crime risk for a specific address inside the radius. It is not directly comparable to the headline figure for an area with very different population density, because no denominator has been applied. And it is not a count of all crimes that occurred — only those that came to police attention and survived the recording process. Under-reporting varies by category and by area, and the headline figure inherits that variance silently.

The category breakdown

Police UK groups recorded crime into 14 standard categories, and the report shows each one as a count alongside its share of the 12-month total. This is the section most readers spend the longest on, because it answers the question that drives most reports in the first place: what kind of crime is happening here.

Two categories cause more confusion than the rest. The first is the distinction between anti-social behaviour and public order, which sound interchangeable but are recorded under different rules and capture different kinds of incident. The second is violence and sexual offences, which is a single combined umbrella in the Police UK schema rather than two separate categories — a count under that label can include a wide spread of incident types, and the total alone cannot be split further from the open data. We have written a dedicated piece on the first distinction at anti-social behaviour vs public order, and we recommend reading it before drawing conclusions from either count.

The trend chart

Below the breakdown, the report includes a small line chart showing total recorded crime in the radius for each of the past 12 months. The line is smoothed lightly to take the edge off month-to-month jitter, but the underlying monthly counts are preserved on hover.

The chart is best read as a slope across the whole window rather than as a sequence of individual points. Month-to-month variation in a small geographic area is largely random — for a half-mile radius in a quiet suburb, a single category may post counts in the low single digits month after month, and the difference between three incidents and seven is dominated by noise rather than signal. A reader who reacts to a single dip or spike will, more often than not, be reacting to nothing. A reader who looks at the slope across all 12 months, and notes whether the line ends near where it started or somewhere meaningfully different, will get the most out of this section.

The AI safety summary

Every report includes a short paragraph generated by a language model that translates the numbers in the report into prose. It is meant to save the reader the work of holding 14 category counts and a trend slope in mind at once.

The summary is not an opinion, a recommendation, or a prediction. The model has been given the statistics and only the statistics — no incident narratives, no police commentary, no neighbourhood context that the figures themselves do not carry. Treat it the way one would treat the executive summary at the top of a longer document: a faithful condensation of what is below it, useful for orientation, not a substitute for reading the underlying figures. We describe the generation process in more detail in how our AI safety summaries are written.

What the report deliberately doesn't show

Several things a reader might reasonably expect to find are absent by design. The report does not predict future crime — we do not run forecasts, and the figures describe what was recorded, not what is likely to happen next. It does not assign a risk level to a specific address inside the radius, because a half-mile circle contains many addresses and the micro-context of each one (a busy junction, a quiet cul-de-sac, a particular building) is not visible in the open data.

The report also does not give a comparison to a national average. We have chosen not to normalise to population density at this granularity because the denominator is poorly defined for an arbitrary radius — residential population, daytime footfall, and visitor numbers all matter, and none of them are cleanly available for a half-mile circle. And the report contains no witness reports, no suspect details, and no information of any kind that is not already in the public Police UK release.

How to use it well

• Treat the report as one input among several. Combine it with a walk-around of the area, a conversation with a neighbour, and the official Police UK site if a full street map is needed.
• Compare like with like. Two postcodes of similar density will produce more honest comparisons than one urban postcode set against one rural one, even if the headline numbers look superficially comparable.
• Re-pull the report monthly if an area is being tracked over time. New Police UK data lands roughly six weeks after the month it covers, so a report run on the same postcode a month later will quietly include a new month at the end of the window.

Where to go next

For more on the data the report is built from, see Police UK's open data: what's in it, what's not. For a worked example of the report read in anger, with two real postcodes set side by side, see comparing two postcodes.