Methodology

LFR Monitor is an independent register of live facial recognition (LFR) deployments by UK police forces, built because no single official source tracks this consistently across forces. Each force publishes its own deployment logs — if it publishes them at all — in its own format, on its own schedule, using its own terminology. This page explains how those logs are turned into one comparable dataset, and where that process involves a judgement call rather than a simple fact.

What data is shown

Each row represents one deployment: one force, one location, one date. Where a force's own log records more detail than others — camera counts, intervention outcomes, alert-confirmation splits — that detail is kept rather than discarded to fit a lowest-common-denominator schema. This means some columns are empty for some forces; that's a reflection of what was actually published, not a scraping error.

How the data is collected

Sources are force-published registers, PDFs, or spreadsheets — see each force's page for direct links to the source documents.
Figures are extracted into per-force canonical datasets, then aggregated into a UK-wide dataset.
Automation is deliberately limited. Several forces' websites use measures that make unattended scraping unreliable, so deployment logs are periodically re-checked and re-verified by hand rather than assumed to update themselves. Every force's page shows when its data was last verified.

Harmonising different forces' terminology

The hardest part of this project isn't collecting the numbers — it's deciding when two forces' different terms mean the same thing. That decision is never made silently. Three real examples from the current dataset:

Case	Decision
Merseyside records `faces_seen_estimate` rather than a directly logged count	Merged into the same "faces scanned" figure as other forces, but flagged on Merseyside's page as an estimate rather than an exact count — the underlying question is the same, the confidence isn't.
Some of South Wales Police's earlier deployments have no recorded faces-scanned figure at all	Treated as a genuine historical gap (SWP began recording this metric partway through its deployment history), not an error — those rows show as blank, and are excluded from totals rather than counted as zero.
The Metropolitan Police's per-deployment "faces scanned" figures were coming out as zero for almost every row — inconsistent with their published annual totals	Traced to a column-mapping bug in the ingest script, not a gap in the Met's own published data: the figure was landing in the wrong field for most 2023 onward deployments. Fixed and re-verified row-by-row against the original source PDFs (see `data-dictionary.md`) — the Met's faces-scanned figure is now published with the same confidence as other verified forces.

Why this matters: a site that quietly merged "estimate" and "exact count" into one number, or published an unverified figure because it looked plausible, would be making the data look more solid than it is. The goal here is the opposite — verified confidently, gaps shown as gaps.

Accuracy and limitations

Forces use different definitions and reporting formats. Fields are kept faithful to their source rather than forced into agreement where the underlying measurement genuinely differs.
Not every force publishes deployment data, and some publish only partial histories. A force showing no data on this site may still be using LFR — it means no published register has been found or verified yet, not that no deployments have happened.
Because full automation isn't reliable against several forces' websites, there will always be a delay between a deployment happening and it appearing here. The "last verified" date on each force's page is the honest measure of how current that force's data is.

Spotted an error, have a source this site is missing, or work for a force and want to correct something? Use the Contact page.