Police.uk publishes street-level crime about seven weeks after each month ends. Looking ahead brings each area on the map up to date and three months beyond: an expected range for the month police.uk has not yet published (October 2026), for this month (November 2026), and for December, January and February. Each range is fixed and dated before we open the police data for that month, then checked against it here, misses included. Over time this page becomes a public record of how far recent police data can see ahead, and where it cannot.
What it is not: intelligence, a prediction about any person or business, or a judgement about a place. It cannot see a change that has not yet reached the police data, such as a store that has just started reporting every theft.
September 2026: the first live check
On 12 November we set a range for every area for September 2026, using police data up to August. We had not opened September's data. When we did, between 13 and 15 November, we checked every area against its range.
| Group | Areas checked | Inside the range | Share |
|---|---|---|---|
| London | 74 | 63 | 85% |
| Rest of the UK | 221 | 184 | 83% |
| All areas | 295 | 247 | 84% |
| Areas with 10 or more offences a month | 198 | 161 | 81% |
The UK total for these areas came to 7,943 against a police figure of 8,074, 1.6% under. Ranges are designed to catch the real figure about eight times in ten, so a share near 80% is the target, not a shortfall. Cumbria and Gloucestershire are not included: police.uk has not published September crime data for either force.
Scorecard by area size, September 2026. "Typical miss" is the median gap between the range's middle and the police figure, as a share of the police figure. "Average lean" is how far the middle sat below (+) or above (−) the police figure on average, in offences.
| Area size (offences a month) | Areas | Inside the range | Typical miss | Average lean | Missed above / below |
|---|---|---|---|---|---|
| Under 10 (not shown on the map) | 97 | 89% | 45% | +0.7 | 11 / 0 |
| 10 to 30 | 107 | 85% | 29% | +1.1 | 13 / 3 |
| 30 to 100 | 82 | 77% | 21% | −0.8 | 9 / 10 |
| 100 or more | 9 | 78% | 15% | +0.9 | 1 / 1 |
| All areas | 295 | 84% | 28% | +0.4 | 34 / 14 |
This scorecard will be published for every month as it is checked. Areas whose boundary is upgraded after their ranges were fixed (seven on 16 November) keep their fixed ranges in the files, but the map withholds them until the next forecast round, because the ranges were set on the old boundary.
Where it missed, and why. Most misses came in above the range, in places that had already jumped in August, where a three-month average reacts slowly. A few were sudden falls. One example: Love Hastings told us in early November that a store in the town centre had started reporting heavily, and that the centre's figures might rise. Our range missed that; the BID saw it coming. That is the case for BIDs sharing their own monthly figures, which arrive weeks before the police data.
The blind test
We also tested the method on the past, to see whether September was typical. For every month from September 2024 to August 2026, we forecast one, two and three months ahead using only the data available at the time. The ranges were set on the first year and then tested blind on the second (September 2025 to August 2026), which played no part in setting them.
| Area size (offences a month) | Inside the range | Typical miss |
|---|---|---|
| 100 or more | 78% | 13% |
| 30 to 100 | 81% | 22% |
| 10 to 30 | 79% | 30% |
| Under 10 (not shown on the map) | 78% | 44% |
We ran the same test further out, because a range for February 2027 is five months beyond the newest police data (September 2026).
| Months beyond the newest police data | Used for | Inside the range | Typical miss (10+ a month) | Typical gap on the total |
|---|---|---|---|---|
| 1 | October 2026 | 78% | 24% | 3% |
| 2 | November 2026 (this month) | 79% | 26% | 5% |
| 3 | December 2026 | 77% | 27% | 8% |
| 4 | January 2027 | 78% | 27% | 9% |
| 5 | February 2027 | 78% | 28% | 10% |
Accuracy barely drops from one to five months ahead, though ranges widen slightly and the national total drifts more the further out we look. That is what makes a range for this month, and the next three, possible despite the seven-week lag. Totals across many areas are much steadier than single areas, because individual swings cancel out: the national total was typically within about 7% in the blind test. Areas with fewer than 10 offences a month are too small for a useful range, so the map says so instead of showing one.
How the range is worked out
- Recent level. The area's last three months of recorded shoplifting, with the latest month counting for half.
- Season. Adjusted by how the same months moved last year across all areas on the map.
- Recent accuracy. Adjusted by how the last six months' expectations compared with what police actually recorded, so the method keeps up when the national trend shifts.
- Range. Set from how far real figures strayed from the expectation in the earlier test year, separately for each area size and for one, two and three months ahead.
The method is version 2. Version 1 used a plain three-month average with no recent-accuracy step. Version 2 was chosen on the blind test year, not on September, and fixed on 15 November. Version 2 also sets December to February. Both versions' October and November ranges are kept and both will be checked, so the record shows whether the change helped.
The rules we hold ourselves to
- Every range is fixed in a dated file before we open that month's police data. The files and their fingerprints are listed below.
- Fixed files are never changed. A change of method gets a new version number and its own file.
- Every check is published: inside or outside the range, and the gap on the total.
- Misses are shown as prominently as hits, with a reason where one is known.
- Ranges are only given for whole areas, never for a single store or street point.
Fixed files
Each file's SHA-256 fingerprint lets anyone confirm it has not been altered since the date shown. The files are on the data download page, after agreeing to the terms of use.
| File | Fixed | What it holds | SHA-256 |
|---|---|---|---|
| forecast_2026-11-12_september.csv | 12 Nov 2026 | September ranges, all areas, version 1 range widths | eef64bb2b9df284a43216a0e3856eda5f7096dfc020324def800b7dd23c304d2 |
| forecast_locked_2026-11-15_pilot_v1.csv | 15 Nov 2026 | September to November, 25 pilot areas, version 1 | 52f8ccca991e1a0c20466c5c9bb9cb86587aedda1d806a9e576a864181173482 |
| forecast_locked_2026-11-15_all_areas_v1.csv | 15 Nov 2026 | September to November, all areas, version 1 with ranges by area size | 6d446840fb7fbc3098283b2c8e9c0130f6c8686e7b13bd6bc2a0adc66f51998c |
| forecast_locked_2026-11-15_all_areas_v2.csv | 15 Nov 2026 | October and November, all areas, version 2 (shown on the map) | 4fd16052bb3bf130bd8f352c12c46636848631ac86019b645412ee08e1753b2e |
| forecast_locked_2026-11-15_all_areas_v2_dec-feb.csv | 15 Nov 2026 | December 2026 to February 2027, all areas, version 2 | 8766d108439d854888e2d569fed770b8873ffe9ae912bf0c0f32441cbcf02b05 |
| scored_2026-09_outside_london.csv | 15 Nov 2026 | September check, areas outside London | 9a535fb4c368f74001e347a3a7cbb6c50f4e1560090edf334a617dd0659e5104 |
| scored_2026-09_london.csv | 15 Nov 2026 | September check, London | 10a2df36f2262e297b850a2b07712f3b75704d26e0dd6c2901840aeb4e5155f8 |
Help it see further
Recent police data cannot see what has not reached it yet. A BID's own monthly incident counts, in police crime categories and with no names, arrive weeks before police.uk publishes. Set against these ranges, they show early when a place is moving and why. If your BID would share a short monthly return, get in touch here or email dan@oxcam.online.
How to cite
UK Business Crime Map, Member Business (crimemap.oxcam.online), data version v66, November 2026. Published under CC BY 4.0. Built on police.uk street-level crime data (Open Government Licence v3.0). If you use these ranges in a report or an AI tool, please keep the credit, and the point that they are expectations with a stated range, not predictions.
