Splink is a probabilistic record-linkage engine built by the UK Ministry of Justice. Its job is to decide when two records describe the same person. This is the list of the governments and agencies that run it, what they run it on, and how far each claim actually stands up.
Every entry below is graded. Verified means the organisation’s own document names Splink. Self-reported means only the Ministry of Justice’s own promotional showcase names it, and the organisation has never confirmed it. Evaluated means they tested it and did not adopt it. Two claims that did not survive checking are named and removed, because leaving them in would be the easy mistake.
National statistics offices. The UK’s Office for National Statistics links the 2021 Census to the Census Coverage Survey and to administrative data. The Australian Bureau of Statistics built its 2025 Person Linkage Spine with Splink, joining Medicare, Centrelink welfare and Tax Office income records across the ever-resident population. Germany’s Destatis ran a Splink method test in preparation for its 2031 register-based census. Statistics Canada tested it on Databricks for census duplicate matching. Uruguay’s INE used it in the 2023 census.
National departments. The Ministry of Justice itself runs it across courts, prisons and probation, linking more than a hundred million records and, in one pilot, pushing Police National Computer numbers to police. NHS England uses it to link health datasets to the national patient spine. The Department for Business and Trade links company and product data with it. The Department for Education matches provider records to learners. The Competition and Markets Authority uses it to identify persons of significant control across companies. The Office for Health Improvement and Disparities uses it to link probation records to addiction-treatment records.
A foreign health ministry. Chile’s Ministry of Health, with University College London, linked the National Immunization Register of about 77.9 million records to the school-enrolment database of about 68 million, to find immunisation gaps among migrant children who have no national ID.
Sub-national. The US state of Indiana runs cross-agency probabilistic linkage of state datasets with it. In Wales, the SeRP linkage centre at Swansea University uses it to build population research cohorts.
The Ministry of Justice publishes a showcase of roughly forty users. Treat it as what it is: a marketing list. It has already been caught overclaiming twice, so nothing on it counts as confirmed until the organisation itself says so. Where only the showcase names the tool, the entry stays here. In the UK that is the UK Health Security Agency (HIV testing linked to health records), Homes England (Land Registry to Ordnance Survey addresses), the Welsh Revenue Authority, the Ministry of Defence (the Veteran Card service), and six English councils using it for debt, children’s services and free-school-meal targeting. Abroad it is the Ontario Ministry of Children, Community and Social Services, Environment and Climate Change Canada, and the 2024 Gambian census, which no Gambian source corroborates. Each of these may well be real. None is confirmed by the organisation named.
Three of the most striking entries on the maintainer’s own showcase collapse on inspection, and each is worth naming precisely because they are the ones a careless write-up would lead with.
UNHCR. The Ministry of Justice lists the UN refugee agency as a Splink user for deduplicating refugee data. UNHCR’s own record-linkage code uses a different tool, fastLink, and no UNHCR document names Splink. On the available evidence the claim is not supported. Removed.
The European Medicines Agency. Listed as using Splink to find duplicate adverse-event reports. The EMA’s own materials describe a terminology-based duplicate-detection method, and the word Splink appears nowhere on its site. Removed.
The US Defense Health Agency. The showcase’s most quotable line, deduplicating military hospital records across some 200 million data points, rests on no source but the showcase itself. No US government document names Splink. A claim that specific, that sensitive, and that thinly sourced does not belong in a list of confirmed users, especially after the two above. Removed.
For completeness: the US Census Bureau evaluated Splink against its own BigMatch tool and did not adopt it, and France’s INSEE recommends it in a methods guide while its own worked examples use other packages. Neither is a user.
The absences are part of the map. Searched and not found, each running its own linkage systems: Statistics New Zealand, the Netherlands, all of the Nordic offices, Italy, Spain, Portugal, Ireland, Switzerland, Austria, Poland, Eurostat and most of central Europe, Brazil, Mexico, Colombia and Peru, South Africa, Kenya and Nigeria, the Gulf and the Levant, India, Singapore, Indonesia, the Philippines and Japan. Scotland and Northern Ireland run their own. Brazil’s CIDACS, often assumed to use it, uses its own engine. The pattern is not universal adoption. It concentrates in the UK and a handful of Anglophone and Western statistical agencies.
Strip out the noise and one ministry’s software now decides identity inside census, tax, welfare, health, justice, trade and immunisation systems across at least four countries’ governments. That decision has an error rate, and it is the one number none of them publish. I asked three of them and none could give it. This map is who is making the join. The open question is how often it is wrong.
Graded 10 September 2026. Verified = the organisation’s own document names Splink; self-reported = only the Ministry of Justice showcase names it. Corrections and additional first-party sources welcome.