Methodology

Last rebuilt: 14 October 2026. Sources dated 1–7 October 2026.

This page exists so that anyone on the board can see exactly how their position was calculated. No part of the operator ranking can be bought. Supplier placement is sold separately and is always labelled as paid.

Sources

Nothing is scraped from operator websites. No API keys are required for any source. Both publishers release monthly, so the board is rebuilt monthly and every page carries the source date.

What counts as a care home

A CQC location is included when its Care home? flag is Y. That gives 14,883 locations, of which 14,764 are active and 119 are registered but dormant. Dormant locations are excluded from all rankings and aggregates.

Locations are classified as nursing (4,482) or residential-only (10,282) from the CQC service-type flags. 101 carry both.

Ratings

Each location contributes an overall rating plus five domain ratings — Safe, Effective, Caring, Responsive, Well-led — mapped to a numeric scale:

Outstanding 4 · Good 3 · Requires improvement 2 · Inadequate 1

Ratings are taken from the location-level assessment. Where a location provides more than one service type, CQC no longer issues a single location-level rating and publishes Not Rated instead; for those we fall back to the service-level rating for the Care Homes population group. 13,617 locations use the location-level rating, 80 use the service-level fallback, and 1,186 have no rating at all. 13,679 of 14,883 locations (91.9%) carry a scoreable rating.

Ratings CQC marks as inherited (carried over from a predecessor registration) are flagged in the data and shown on the board, because they describe a different operator's inspection.

Inspection dates range from April 2016 to September 2026. 40% of rated homes were last inspected more than five years ago. The board therefore reports the age of each rating rather than presenting ratings as current fact.

Operator identity

This is the part most likely to be wrong, so it is documented in full.

CQC records a Brand for about 42% of care homes. That field sometimes names the investor or landlord rather than the trading operator. The clearest case is BRAND Welltower — a US healthcare REIT whose 642 CQC-branded homes contain HC-One, Care UK Community Partnerships, Aria Healthcare, Ideal Carehomes, Danforth Care and others. Treating that as one operator would have produced a board topped by a landlord that trades under none of those names, and would have buried HC-One at one home.

So operator identity is resolved in this order:

1. Brands listed as containerBrands in data/overrides.json are dissolved and their homes re-grouped by provider company name against an ordered list of providerAliases.
2. Remaining CQC brands are used as-is, after applying brandMerges (which folds Care UK Community Partnerships into Care UK).
3. Homes with no brand are grouped by provider company.

Companies that fall inside a dissolved container brand and match no alias are grouped as Welltower (unattributed opcos), excluded from the board, and written to data/out/review-queue.json for a human to name. That currently covers 36 companies, 256 homes and 17,696 beds (3.6% of national beds) — mostly entities named WT ... Opco, Scarborough Hall Limited, Willow Tower and Redwood Tower, whose trading brand cannot be determined from open data.

Operators using one special-purpose company per home (Exemplar: 58 homes, 58 companies) are normal and are not dissolved.

This layer is a judgement call, not a fact. It is reviewed before each publish.

Duplicate registrations

When a home moves between legal entities — routine inside large groups and constant during acquisitions — CQC issues a new location ID, and the previous registration frequently stays active in the published file. Both records then describe one building with one set of beds.

495 such pairs were found, double-counting 32,156 beds — 6.6% of the national total. They cluster in acquisitive operators (Care UK 168, HC-One 45, Barchester 41), so the effect on the ranking was not random: because scale is 28% of the index, it put the wrong operator in first place.

Records are grouped on name, local authority and bed count. The most recently registered is kept, because that is the entity operating the home today; where it has no rating of its own the predecessor rating is inherited, since the inspection happened at the same building. Every merge is logged in data/out/merged-locations.json.

Company data

Provider companies are matched to the Companies House bulk snapshot on company number, zero-padded to eight characters. 6,310 of 6,343 numbers matched (99.5%), covering 13,535 of 14,883 homes (90.9%).

From that we take incorporation date (for years trading), company status, accounts filing category (a coarse size band, not turnover), and the count of outstanding mortgage charges. Four housing associations carry a 01/01/1900 placeholder incorporation date rather than a real one; these are treated as unknown rather than as 126-year-old companies.

"In distress" means a company status of Liquidation, In Administration, Voluntary Arrangement, Receivership, or Active – Proposal to Strike off.

Board eligibility

An operator qualifies for the board if it has:

• 3 or more care homes, and
• 60 or more registered beds, and
• 2 or more homes with a scoreable rating, and
• a resolved trading identity (not an unattributed opco group)

543 operators qualify. The published board is the top 500 of those, covering 286,905 of 488,547 registered beds — 58.7% of the England market. The 43 that qualify but fall below the cut keep their score and are promoted automatically if they rise in next month's rebuild.

The thresholds are about comparability, not size. An operator needs enough homes, and enough *rated* homes, for an average to carry any meaning: an earlier draft of this board ranked a five-home operator with 33 beds and three perfect inspections at number two, which told the reader nothing.

Operators that do not make the board are not hidden. Their homes appear in full in the local-authority directory, with the same bed counts, ratings and inspection dates as everyone else. Only the ranked 500 get a profile page.

Index Score

Each component is converted to a percentile rank within the eligible pool, then weighted. Percentiles rather than raw values, so that no single metric with a long tail can dominate the result.

Quality is bed-weighted so that a group's large homes count for more than its small ones. Freshness is included because an operator whose ratings are all six years old is telling you less than one inspected last year — it measures the age of the evidence, not the operator.

Where a component is missing, the remaining weights are renormalised rather than scoring the operator zero.

Scores are published to one decimal place, so ties happen. Where two operators tie, the one with more registered beds ranks higher; if they tie on beds too, they are ordered alphabetically. That rule is fixed, so the board is identical on every rebuild of the same source data.

Ranking an individual home

The operator index is the wrong tool for a single building. It weights scale at 28%, which is right when comparing businesses and wrong when comparing homes: a 90-bed home is not a better home than a 30-bed one, and someone reading a town page is choosing a building.

So the home score excludes scale entirely:

The weakest-domain term is the part worth explaining. A home rated Good overall but Requires Improvement on Safe is not the same proposition as one that is Good across the board, and 1,459 homes are in exactly that position. No consumer directory surfaces it.

Recency is percentile-ranked rather than scored on a fixed window. An earlier version scored anything older than six years as zero, and because 40% of England has not been inspected in five, 32% of homes landed on an identical total and the ordering within a town was effectively arbitrary. Percentiles spread the one continuous variable in a dataset where 71% of homes share the same coarse rating; the largest tie group is now 0.6%.

Position on a town page cannot be bought, on the same terms as the operator board. No care home has been asked to pay and none can.

Town pages and geography

Postal town is the unit people search — "care homes in Reading". Within a town, postcode district is what a phrase like "East Reading" actually means: RG6 is Earley, RG4 is Caversham. Districts appear as *sections* of the town page rather than pages of their own. Publishing 771 district pages with five homes each, all at once, from a domain days old is the shape that Google's helpful-content system exists to catch, and it would split one town's authority across near-duplicates competing with each other. A district can be promoted to its own page later if it shows real demand.

CQC spells towns inconsistently — Stoke-on-Trent appears five ways, and 45 towns have more than one spelling. Towns are grouped on a normalised key and displayed under the best-formed spelling actually in use.

Known limitations

• Bed counts are registered capacity, not occupancy. CQC does not publish occupancy. Nothing here should be read as a revenue proxy.
• Ratings are stale across much of the sector. 40% of rated homes were last inspected over five years ago. The board shows rating age for this reason.
• Accounts category is a filing band, not turnover. Real revenue would need per-company accounts, which the bulk snapshot does not include.
• The operator layer needs ongoing human review. See the review queue above.
• England only, by design. CQC regulates England. Scotland (Care Inspectorate), Wales (CIW) and Northern Ireland (RQIA) publish separately and are deliberately out of scope, so the board is titled England throughout.

Corrections

Any operator can request a correction or removal. Corrections to factual data are passed upstream to CQC or Companies House where the error originates, and annotated here in the meantime. Removal requests are honoured within one working day.