Home › Method
How this is produced
Every figure on this site comes from accounts that companies filed themselves. This page says exactly how they are read, so that if a number here is wrong you have enough to prove it.
The source
Companies House publishes the accounts registered with it as open data, one archive per working day, at download.companieshouse.gov.uk. A typical day is about ten thousand filings and 80MB. We download it each morning, read every filing, and join each one to the Companies House company register for the name, the industry and the registered office.
So far that is 90 publishing days and 1,266,270 sets of accounts. There is no back catalogue beyond a few months, so a day not collected is a day lost for good. That is why the collection runs whether or not anything is published.
What the dates mean
Each daily report is dated by the morning Companies House published the file, not by the day any company filed. Companies House adds a new file each morning from Tuesday to Saturday. In its own words, each file "will only contain the accounts data registered on the previous day, with the exception of Tuesday's file which will contain data registered on the previous Saturday, Sunday and Monday". So the report for Thursday 22 October covers accounts registered on Wednesday 21 October, and every report says which registration day or days its file covers. There is no file on a Sunday or a Monday.
The reports from 20 June to 11 September 2026 were compiled together when the site launched on 12 September. They are now summarised as monthly pages, and each old daily address redirects to its month. From 12 September each report was published on the day of its file, or the morning after.
How each daily report is written
A pipeline writes the report from the day's figures, and the lead is whichever movement was largest that day: the number of accounts, the share in deficit, the share weaker than a year earlier, or an industry far from its own record. Every comparison is like with like. The share in deficit is compared with the share in deficit on the previous publishing day, the same day a week earlier and the average of all earlier days; the share weaker than a year earlier is compared only with its own average. An industry or town is compared with its own earlier filings, allowing for how many filed, and only earlier days are ever used.
Each report also lists the day's largest employers by the average headcount their accounts state. That is the only reason a company is named. Headcounts that are almost certainly keying errors are left out: a figure between 2015 and 2030 (usually a year), a figure of more than 5,000 that the company's own prior year does not support, and any figure of 500 or more not backed by at least 2,000 pounds of assets per employee.
What is read from each filing
Accounts are filed as iXBRL, which means the figures carry machine-readable tags. We take net assets or liabilities, equity, current assets, creditors and the average employee count, for both the current year and the prior year that every filing includes alongside it. That prior year is what lets a single day of filings show a direction of travel.
Coverage is good but not total: about 96 per cent of filings tag equity and about 81 per cent tag net assets. Turnover appears in roughly 1 per cent, because small companies file abridged accounts with no profit and loss account. We do not report turnover for that reason.
Two bugs worth admitting
Both were caught before anything was published, and both would have produced confident wrong numbers.
The namespace bug. The tag prefix in these filings is local to each document: core, ns5, frs-core, uk-core and others all appear within a single day. Keying on one prefix returned 13 per cent coverage, which looked like most companies had simply not reported a balance sheet. Matching on the tag name instead took it to 96 per cent.
The scale bug. Accounts presented in thousands carry a scale attribute, and filing software routinely applies it to employee counts as well as to money. A company with 203 staff was read as having 203,000. Honouring scale only for monetary figures cut one day's employee total from 350,325 to 53,412.
There is one error we cannot fix: filers occasionally type the year into the employees box, so a company reporting 2,023 staff is usually reporting 2023. It is one reason we do not build a story around any single company's figures.
What a deficit means, and what it does not
When we say a filing is in deficit we mean one thing: the accounts reported liabilities larger than assets at the balance sheet date. That is not insolvency, and we never say it is. A company funded by its parent, by a director loan, or one still carrying early losses forward can report negative net assets for years while paying everything it owes on time.
It is a fact about a filing, not a verdict on a business. That is why these reports deal in aggregates and industries rather than naming companies as failures.
What this cannot tell you
- Companies House does not verify what is filed with it. Everything here is what companies reported about themselves.
- A registered office is an address for service, often an accountant, so the geographic split is a weaker signal than the industry split and we say so.
- SIC codes are self selected, capped at four, and never audited. They are reliable in bulk and unreliable one company at a time.
- Publication dates are not financial year ends. A single day's file covers many different accounting periods, which each report shows.
- There are no market valuations here. This is what companies file, not what they are worth.
Corrections
If a number here is wrong, write to editor@capnews.co.uk with the date and the figure. We will check it against the filing and, if we got it wrong, correct it and say what changed. If your own filed accounts are wrong, that correction has to be made with Companies House, and it will reach us in a later day of data.
Source data is Crown copyright, published under the Open Government Licence v3.0.