Sources and method

This page exists so that nothing on this site has to be taken on trust. Every figure is traceable to a named publication with a date, and the rules below describe what we do with those figures and what we refuse to do with them.

If you find an error, we want to know. See corrections.

What this site is, and is not

It is an explainer. It takes official statistics that are already published and tries to make them readable without making them wrong.

It is not a source of original statistics. Every figure it starts from was published by somebody else, and where our figure and theirs differ, theirs is right and ours is a bug. Some figures here are worked out from those rather than taken from them, most often a share this site calculates because no publisher prints it. Each one is graded calculated, says so where it appears, and shows what was divided by what. That is the only original arithmetic on the site, and it is labelled rather than hidden inside a number that looks taken.

It is not a live system. Figures are the latest published at the time of the last update, not a real-time count. There are no live counters on this site, deliberately.

It is not neutral about statistical misuse, though it takes no position on what immigration policy should be. Where a claim misuses a definition, a denominator or a time period, we say so, and we apply that test in every direction. The common claims page says how claims are chosen.

Where the figures come from

Official statistics are the primary source throughout. We use independent research bodies for context, and for periods official sources no longer cover.

Home Office: Immigration system statisticsHome Office

What it coversAsylum applications, decisions, backlog and support; visas granted by route (work, study, family); settlement and citizenship; safe and legal routes; small boats and irregular migration; and returns, including enforced, voluntary and port refusals.

UpdatedQuarterly (latest: year ending June 2026 for asylum, returns, visas, settlement and citizenship)

Home Office: Annual Report and AccountsHome Office

What it coversDepartmental outturn, including asylum support spend and how much of it went on hotels, and the visa, immigration and health surcharge income the department collects. The report is annual; the hotel figure is not. The 2025-26 edition, HC 440, was opened on 10 August 2026: it drops the hotel total and the per-day average entirely, and restates the income figures, so records from this family sit on two editions on purpose and each says which and why.

UpdatedAnnual (latest: 2025-26, HC 440, July 2026)

ONS: Long-term international migrationOffice for National Statistics

What it coversNet migration, immigration and emigration estimates, and reason-for-migration breakdowns. The headline net migration measure.

UpdatedTwice yearly (provisional; official statistics in development)

ONS: Population estimatesOffice for National Statistics

What it coversUK mid-year population, used as a denominator for shares and per-person context.

UpdatedAnnual

Census 2021/22ONS (England, Wales, Northern Ireland) and National Records of Scotland

What it coversForeign-born population, country of birth, nationality, and regional distribution, and the basis the foreign-born figures on this site ultimately rest on, the ONS survey-based series having been discontinued in October 2022. NO RECORD CITES THIS ENTRY DIRECTLY, and it is listed because a reader is owed the underlying source: the figures are cited to the Migration Observatory briefing and the ONS estimate that read the census, because those are what this site fetched and quoted, and no census release publishes a UK-wide foreign-born share at all.

UpdatedEvery 10 years (last: 2021 for E&W/NI, 2022 for Scotland)

The Migration Observatory (University of Oxford)University of Oxford

What it coversNeutral analysis and briefings on the migrant population, labour market, fiscal impact, and long-run trends, drawing on ONS, Census and Home Office data.

UpdatedOngoing

National Audit OfficeNational Audit Office

What it coversCost and value-for-money of the asylum system and asylum accommodation contracts.

UpdatedOccasional (latest: An analysis of the asylum system, December 2025)

Independent Chief Inspector of Borders and ImmigrationICIBI (independent statutory inspector; reports laid before Parliament)

What it coversInspections of Home Office immigration and asylum operations, including contingency asylum accommodation; relays Home Office operational and cost data.

UpdatedPeriodic inspection reports (irregular)

Office for Budget ResponsibilityOBR

What it coversFiscal impact of migration in forecasts and the Fiscal Risks and Sustainability report (estimates and scenarios, clearly labelled).

UpdatedAt fiscal events and in periodic reports

Migration Advisory CommitteeMAC (advises the Home Office)

What it coversSalary thresholds, shortage occupations, and the economic and fiscal evidence on work migration routes.

UpdatedAnnual report plus commissioned reviews

Tribunals statistics quarterly (HMCTS / Ministry of Justice)Ministry of Justice

What it coversAsylum and immigration appeals: lodged, decided, outstanding, success rates, and waiting times.

UpdatedQuarterly (latest: April to June 2026)

House of Commons Library research briefingsUK Parliament

What it coversPlain-English summaries of asylum statistics, small boat crossings, NHS overseas staff, and immigration fees, drawing on the primary sources above. No figure a reader meets on this site comes from it at present. It stays listed because this site holds one figure from it in reserve, its share of NHS staff who are not British nationals.

UpdatedOngoing

Skills for Care: state of the adult social care workforceSkills for Care

What it coversNationality composition of the adult social care workforce in England.

UpdatedAnnual

Office for Statistics Regulation: guidance on migration and asylum statisticsOffice for Statistics Regulation

What it coversGuidance on responsible interpretation of migration and asylum statistics: which comparisons are valid, and where published figures are commonly misread. Used for methodology and caveat design rather than as a source of figures.

UpdatedOccasional

ONS: EMP06, employment by country of birth and nationalityOffice for National Statistics

What it coversEmployment levels and rates for people aged 16 to 64 by country of birth and by nationality, from the Labour Force Survey, not seasonally adjusted. The primary table behind the employment-rate comparisons between UK-born and migrant residents.

UpdatedQuarterly (official statistics in development)

On the Migration Observatory and Skills for Care. These are not official statistics producers, and figures attributed to them are marked as such. We use them where they add methodological context, or where they hold the most reliable available figure for something the official series no longer covers, the foreign-born population being the main example.

On the Office for Statistics Regulation. No figure on this site comes from OSR. Its guidance informs how we present and caveat the figures that do.

How every figure is recorded

Figures on this site are not loose numbers typed into sentences. Each one is a record carrying its own provenance, and the site is built from those records.

Every figure carries: what it counts, its value and unit, the end date of the period it covers, a human-readable period label, its geography, the publishing body and publication title, a link to the source, the publication date, the date we last checked it, notes on what it does and does not show, and a confidence level.

Three consequences worth stating plainly:

• A figure exists in one place only. Nothing on this site copies a value from somewhere else on this site. Pages reference the record, and so do the charts: a bar names the record it draws rather than carrying a number of its own. This is why you will not find two different figures for the same measure on different pages, a failure common enough on data sites to be worth designing against.
• Prose cites records, it does not restate them. Where a number appears in a sentence on this site, the build inserts it from the record. Updating a figure updates every sentence that quotes it, chart summaries included. One kind of number is written out instead, and it is declared rather than tolerated: a figure that was true when it was published and has since been superseded, where the point of the sentence is that it moved. The two earlier net migration estimates under Estimates get revised are exactly that. A record holds what a figure is now, so it cannot hold what a figure used to be; each frozen number is listed in its own page's front matter with the reason it is frozen, and the checks refuse an undeclared one.
• Checks enforce these rules, rather than a person remembering them. A figure missing its source, a date that does not fall within the period it claims, a range presented as a single number, two versions of the same measure, a chart bar with a number typed into it, or a current value written out longhand anywhere in a page will each stop the change being published.

Confidence levels

Every figure is graded. The grades are this site's own judgement about how far a figure can be relied on, not the publisher's, and one of them, "calculated", exists to mark a figure this site derived rather than read off a release. The grades mean:

official

Taken directly from an official published release.

provisional

Official but flagged by the publisher as subject to revision (e.g. ONS net migration, operational small-boat data).

estimated

An estimate, interpolation, scenario, or pre-2021 figure on a non-comparable methodology, not a current published official figure.

calculated

Derived by this project from other figures.

What the checks do not establish

Mostly they are about where a number comes from, not about whether the sentence around it is right, and the last of these is not about numbers at all. Stated because a check trusted for more than it does is worse than no check at all:

• A description of a figure is not checked against the figure. A chart summary saying a series rose when it fell would build cleanly, and so would one that quotes the right number against the wrong year. Reading, not tooling, has found the false summaries on this site.
• Values under 100 are reported for review rather than refused. Too many measures share a small value for a match to mean much, so those are flagged on every build and looked at, not blocked.
• The scan for written-out figures reads one set of forms, not every form. It sees a number in digits, grouped with commas, or followed by the word "million" or "billion". It does not see one spaced as "2 200 000", spelled as "two million", abbreviated to "bn", or scaled by any other word. Since 6 August 2026 it reads one front-matter field, the short answer a claim page prints as its answer, which until then was unread and was carrying live figures frozen as text on two pages. The rest of the front matter it still does not read. The rule above holds for what the scan can see; beyond that it rests on editing.
• A correction made inside an edition can pass unseen. Where a publisher revises a table but keeps the same release under the same address, nothing here re-reads it. One publisher's corrections page is watched, and the match is made against the table identifiers a record declares, so a note naming a corrected table by its title instead does not match. The other publishers this site cites announce corrections somewhere this site does not look. This is the limit most likely to leave a wrong number on a page rather than an unchecked one, and it is why the citation block under each page prints the date every source behind it was last read.
• A figure written into a chart's summary sentence cannot be traced back out of the page. Every other figure here carries the id of its record in the page's own markup. That is how the build confirms that the figures a reader meets and the figures this site counts as published are the same set, and it is how each page works out the date it prints at its foot. A chart summary is composed as a single sentence and then handed to a screen reader as the chart's description, so an id cannot be attached to a number inside it without either showing the markup to that reader or keeping the same sentence in two versions. Neither is worth doing to make a check easier. Those figures are recorded, sourced and evidenced like every other one; what is not checked is that they reached the page, and a page whose only figures arrived that way would print no line at its foot saying when its figures were last checked.
• No real screen reader has been run over these pages. The accessibility checks here are an automated audit against WCAG 2.2 AA on every page, plus reading the accessibility tree that assistive technology actually consumes. That is not the same as someone listening to a page with VoiceOver or NVDA, and it is the gap on this list most likely to matter to a reader rather than to an editor.

Caveats that apply across the whole site

These are the traps in UK migration statistics generally, not quirks of this site. They are the reason two honest sources can appear to contradict each other.

1. People vs cases: the Home Office headlines 'people' (main applicants plus dependants) for asylum, while 'applications' and 'cases' count main applicants only. 85,891 people claimed asylum (year ending June 2026), relating to 69,848 main-applicant applications.
2. Reference periods differ, though less than they did. Asylum, returns, visas, citizenship and settlement are all now on year ending June 2026. ONS net migration, immigration and emigration remain on year ending December 2025, because ONS publishes on a different cycle. The asylum applications and backlog timeseries are on calendar years and end-December stocks respectively. Do not compare a year-ending-June figure with a year-ending-December one and call the difference a change.
3. Net migration methodology: ONS now publishes ONE continuous series on its 'new approach' from 2012 to 2025, INCLUDING 2020. The older framing is out of date. It ran the International Passenger Survey to 2019, switched to administrative data from 2021, and treated 2020 as unpublishable. ONS's November 2023 back series supersedes any series previously published for 2012 to 2021, and the old dataset is now titled 'Discontinued after 2019'. The comparability break ONS maintains today is at JUNE 2021, not 2020: comparisons across it should be treated with caution because the methods differ. ONS also states that you cannot append the latest estimates to a series from an earlier release. Use the full series from the current release. Revisions are large: the 2022 estimate is now substantially higher than when it was first published.
4. Foreign-born totals come in two vintages. The census-based figure is the 2021/22 Census (16% / 10.7 million). ONS discontinued its annual survey-based series in October 2022, then introduced a rolled-forward method and published a provisional mid-2024 estimate of 13,115,000 people, about 18.9% of the population. That estimate is an official statistic in development, ONS publishes no percentage, and there is no 2025 or 2026 figure.
5. Asylum work-in-progress (total casework backlog) is stale: the only complete figure is 224,742 cases as at June 2024; the wider breakdown was suspended pending a new caseworking system. The current 'awaiting initial decision' figure (40,168 people) is separate.
6. Fiscal impact is contested: estimates depend on method and assumptions, and across the pre-Brexit static studies summarised by the Migration Observatory they are consistently within about plus or minus 1% of GDP. The OBR £297,000 figure is a stylised single-case scenario, not a typical or average outcome.
7. ONS Labour Force Survey reliability: employment-rate figures by country of birth have known sample-size and response-rate problems since 2020 and should be treated as provisional.
8. Returns count EVENTS, not people: one individual returned more than once is counted more than once. Refused entry at port (18,646 in the year ending June 2026) is counted separately from the enforced-plus-voluntary returns total and must not be added to it.
9. Gross flows are rounded independently of net migration, so they do not always reconcile exactly. Year ending December 2024: immigration 1,012,000 minus emigration 680,000 gives 332,000 against a published net figure of 331,000. This is rounding to the nearest thousand, not an error, but do not present the subtraction as if it derives the published figure.

The two that catch people most often are worth putting in plain terms.

People and cases are different counts. In the year ending June 2026 there were 85,891 people claiming asylum and 69,848 main-applicant applications. Both are correct; they count different things. Dividing one by the other, or comparing a figure on one basis with a figure on the other, produces answers wrong by roughly a fifth, and the error does not show unless you check the basis. Every figure on this site states its basis.

Estimates get revised, sometimes heavily. Net migration for the year ending December 2024 was first published at 431,000, then about 345,000, and now stands at 331,000. The estimate changed because ONS brought in new data and methods, not because the period itself did. When you see a revision, that is usually the statistics system working rather than failing.

Reference periods do not line up

Different measures cover different periods, because different bodies publish on different cycles, and not every period is a twelve-month window. At the last update:

Comparing a year-ending-June figure with a year-ending-December one and calling the difference a change is invalid. Where this site puts two figures near each other, it labels the period on both.

How often this is updated

Most of this site is driven by three releases, and the commitment is to update within one month of each of them publishing new figures:

The other 14 published figures come from six publishers outside those three cycles. Some publish annually and some publish when they publish, and none of them carries a promised update schedule here: the National Audit Office, the Migration Observatory, the Office for Budget Responsibility, the Independent Chief Inspector of Borders and Immigration, ONS population estimates, and the Home Office's annual report and accounts. The count above is derived and this list is typed, so they are two homes for one fact: the House of Commons Library stood here until 6 August 2026 and sources nothing a reader meets, which is what made the list say seven where the count said six. Those are re-checked when they publish, and no schedule is promised for them, because a schedule nobody can predict is not a commitment.

One month rather than something tighter because this is one person working unfunded, and a target that is missed is worse than a slower one that is met. The most recent full cycle took twenty-seven days.

Every page shows the date it was last reviewed, and every figure on the home page shows the date it was last checked against its source. If a figure has not been checked within its source's publishing cycle, this site is late, and you should treat the figures as superseded rather than assume nothing has changed.

Before anything is published, an automated check ages every figure against its own source's cycle and reports the ones that are overdue. That runs weekly whether or not anyone is working on the site.

Updating is deliberate, and part of it is now assisted. A second weekly check watches the publishers this site cites and opens an issue when one puts out a new edition, or corrects a table a figure here is read from. It reads publication pages and never touches a figure.

When an update is then made, an assistant may draft it: reading the release, proposing which records move, and opening a pull request. It cannot publish. Every figure whose value changes must carry a quotation from the source that contains the new number, and a check refuses the change if it does not. That catches an invented figure rather than a misread one, which is why a person still reads the source. Merging is a person's decision, and this site's main branch refuses a change that has not passed the checks set to block one. Some checks here report rather than block, and this site's documentation says which.

A whole series goes back to a person on purpose, which covers two of the three regular releases rather than one, because a series is easier to get subtly wrong than a single value.

There is still no automated pipeline pulling numbers straight onto the site, and no figure appears here because a model asserted it.

What this has now been through: on 18 and 19 August 2026 the whole path ran against a real release for the first time. ONS republished the dataset behind two figures in this site's data layer, the release check reported both as citing a superseded edition, each was re-read against the source and updated by hand, and the change was reviewed and merged. One part of it was not exercised and is worth saying so: the check was run by hand on the morning of the release rather than raising its own issue, which it does weekly, so the step where this site is told rather than goes looking is still proved only in its parts. What that update found, including what it corrected in its own notes, is in the changelog.

Corrections

A site that corrects other people's use of statistics has to be visibly better than they are at handling its own mistakes.

• Every change to a published figure is recorded in a public changelog, with the reason and the source.
• Where a revision changes the substance of a claim, the claim carries a dated correction note at the top of the page. It is not silently edited.
• Where a revision changes a number but not the conclusion, the figure updates and the changelog records it.
• Every claim shows when it was last reviewed. A claim unreviewed for twelve months is taken down until it has been.

If something here is wrong, please say so. Tell us which figure, and what you think it should be: open an issue. This site handles corrections in public, so the objection and the answer stay attached to the figure rather than sitting in somebody's inbox. That does mean a correction needs a GitHub account, which is free but is a step. Until 10 August 2026 this page offered an email address instead and said a correction should not depend on having an account; the address was withdrawn, and the condition it existed to avoid now applies. It is a real cost and it is stated rather than dropped.

What this site does not cover

Stating this openly, because silence about scope is indistinguishable from bias.

Crime and migration. Not covered. The statistics are genuinely difficult: nationality is recorded inconsistently across the criminal justice system, the right denominator is contested because migrant and non-migrant populations differ in age and sex profile, and the honest answer to most versions of the question is that the published data cannot settle it. We would rather say that than publish a number we cannot stand behind.

Local authority figures. Not yet covered. Local data needs per-capita denominators and careful handling, and doing it badly produces something closer to a targeting tool than an explainer.

Returns and removals. Covered since 10 August 2026, at returns. This entry previously said the subject was not covered while the data layer held the figures and two of them already reached a reader with no page explaining either, which is the same silence as not holding them. The page does what this entry said covering it would take: it sets out what a return counts, that the counts are of events rather than people, and why the categories do not nest, rather than placing a number beside the asylum figures.

What immigration policy should be. Never covered. That is a question about values, and this site has no standing to answer it.

How this site was built

The figures were assembled in two stages. An initial research pass in June 2026 gathered figures and sources. A verification pass in July 2026 checked them against primary publications, quoting the specific sentence or table cell behind each value it covered. Where a figure has been checked that way, the quotation is recorded in data/evidence/ and a check refuses a changed value whose quotation does not contain it. Every figure here now has one. The backfill finished on 6 August 2026, and since then the check re-reads every quotation on file on every run rather than only the ones a change has touched, so an entry cannot be written once and never asked about again.

That verification found real errors, which is the point of doing it. One long-run series had to be rebuilt entirely: it had silently mixed three different vintages of the same ONS statistic, and one value in it matched no figure ONS had ever published. It was withdrawn and replaced rather than patched.

We used AI in producing this site, and it is right that you know that. It helped with research, drafting and checking. It is not a substitute for verification, and it is exactly why the verification rules above exist: every figure requires a fetched source and a quoted value, anything unverifiable is left out rather than filled in, and the automated checks exist to catch what review misses. No figure appears here because a model asserted it.

Reuse

Most figures here are Crown copyright, published under the Open Government Licence v3.0.

Contains public sector information licensed under the Open Government Licence v3.0.

That covers the Office for National Statistics, the Home Office, the National Audit Office, the Office for Budget Responsibility, the Migration Advisory Committee and HM Courts and Tribunals Service, and the House of Commons Library under the Open Parliament Licence, whose terms are materially equivalent for this use. Two publishers here are not Crown copyright: figures from the Migration Observatory (University of Oxford) and from Skills for Care are short factual extracts reproduced with attribution, their own terms apply, and you should check the source before redistributing those in bulk.

If you reuse them, keep the source and period attached. A figure stripped of its period, its basis and its source is the raw material for exactly the confusion this site exists to reduce.

The site's code and original writing are under the MIT Licence, which covers the prose on these pages, the build code, the documentation and the writing carried inside the data files: the card paragraphs, the notes and the caveats are this project's words rather than the publishers'. The figures themselves are not covered by it; they are under the paragraphs above. The full text of both halves is in the repository's LICENCE file.

The underlying data files, including the source of every figure, are public.

They are here, and they are the same files this site builds from rather than an export of them. Every figure on the site comes from one of these records, and each record carries its own source URL, publication date, the date we last checked it and its confidence grade.

• asylum.json 37 figures
• asylumApplicationsTimeseries.json 16 points
• asylumBacklogTimeseries.json 32 points
• dashboard.json 8 cards
• fiscal.json 18 figures
• meta.json
• migration.json 28 figures
• migrationFlowsTimeseries.json 28 points
• netMigrationTimeseries.json 24 points
• population.json 17 figures
• sources.json 15 sources

Each file is served as JSON at the address shown. data/evidence/ sits alongside them and holds the quote behind every figure that has one, one file per publisher release, with its own README explaining the format.

This page was last reviewed on 19 August 2026. Its figures are not a live count; every one was checked against its source on or after 11 August 2026.