commvita
Connected care platform
Architecture

Send the question, not the data

England's health and care system is thousands of separate legal entities, and that separation is where accountability lives. This is the blueprint for joining it up without dissolving it — what federates at each of the four layers, what it costs, and an honest account of which parts are running today.

Live vs demonstrated: Live — real, API-backed platform logic (wired end-to-end today) Demonstrated — representative control surface with seeded data / illustrative UI mock-up

1The estate is the constraint, and it is not going away

Every proposal to join up health and care data eventually meets the same wall. England's health and care system is not one organisation. It is thousands of separate legal entities, and the separation is load-bearing: it is where accountability sits.

~6,150GP practices~207NHS trusts36Integrated care boards153Upper-tier local authoritiesEach is a separate legal entityits own data controller · its own regulator relationship · its own liabilityThat is not an accident of history to be tidied away. It is the accountability structure of the system.An architecture that requires it to be dissolved is not an architecture — it is a precondition nobody can meet.plus several thousand independent, private and voluntary-sector providers
Indicative counts, August 2026. What matters is the order of magnitude, not the last digit — there is no plausible reform that turns this into one organisation.
And the boundaries are expected to move again. The Cabinet statement of 31 July 2026 sets out an intent to align key public service geographies with the new strategic authorities, naming regional police forces, fire and rescue services and integrated care boards as aligning to strategic authority boundaries by the end of the Parliament — in some cases to more than one such authority — with a white paper on the alignment of public service geographies due in the autumn. Nothing here depends on how that lands. An architecture that assumes a fixed organisational map is wrong within a year, so commvita holds the map as sourced, effective-dated configuration rather than as structure: a boundary change is a data change, not a migration.

2Two ways to join it up, and only one of them survives contact

Centralise — copy the data to the middleOne controller inherits everyone else's liabilityCentral copya new controller, a new targetOrg 1Org 2Org 3Org 4Org 5Org 6every arrow is a data-sharing agreement, a DPIA and a transferand the copy is stale the moment it landsFederate — send the question, not the dataEach org stays its own controller; only answers moveQueryaggregate answer, k≥5Org 1controllerOrg 2controllerOrg 3controllerOrg 4controllerOrg 5controllerOrg 6controller0 rows transferred · counts under 5 suppressedthe answer is computed where the data already lives
The difference is who becomes the controller. Copying every organisation's data to the middle creates a new controller holding other people's liability, a new target, and a copy that is out of date the moment it lands. Federating leaves each organisation the controller of its own record and moves only the answer.

commvita is built the second way. That is a design commitment with consequences, not a slogan, and the rest of this document is the evidence for it.

3Four layers, each federating separately

“Federated” is not one thing. Identity, record, query and governance each have to federate on their own terms, and a platform that does one of them is not federated.

FOUR LAYERS — EACH FEDERATES SEPARATELYIdentityResolve one person across many organisations without merging their records/empi · /nrl-connectorRecordA shared view assembled from contributing systems, not a replacement for them/single-patient-record · /nhrQueryAsk a question of many sites; receive an aggregate, never rows/cohort-discovery · /federated-queryGovernanceWho may ask, of whom, on what basis — and an audit that survives the answer/federated-governance · /ig-spine
Each layer can be adopted without the others. An organisation can federate identity and keep its own record; it can join a query network without contributing to a shared record.
01

Identity — resolve, do not merge

A master patient index resolves a person across sources and identifier schemes, rather than creating one merged record that some organisation then has to own. A national record locator points at where a document lives instead of copying it.

/empi/nrl-connectorLive
02

Record — a view, not a replacement

The shared record is assembled from contributing systems and is explicit that the source systems remain the systems of record. Care homes and community providers contribute and read through the same contract as an acute trust.

/single-patient-record/nhr/care-homes-shared-record/prsb-complianceDemonstrated
03

Query — the answer moves, the data does not

A cohort question is executed at each site and returns an aggregate. Counts below five are suppressed, so a small cohort cannot be narrowed to a person. This is the layer that makes multi-site research and population analysis possible without a transfer.

/cohort-discovery/federated-queryGET /federated-query/nodesDemonstrated
04

Governance — who may ask, and what survives the answer

Data sharing partnership agreements with their signatories and review dates; a Caldicott queue for non-direct-care access; cross-controller notification when a non-controller reads; and an append-only hash-chained access log re-verified whenever it is read.

/dspa-manager/ig-spine/legitimate-access-log/federated-governanceLive

4What makes it real rather than asserted

Scope is computed, not copied

Live

A person's reach is resolved at sign-in from their own organisations, anything delegated to them and any group they belong to. No organisation's boundary is enforced by hiding a menu — the same check applies to a typed URL.

Where it lives effective_scope() · /rbac-matrix

The person can object, and it propagates

Demonstrated

Sharing preferences are held per person, including partial objection by category, and are expressed as a standard consent resource so they travel with the query rather than being re-implemented at each site.

Where it lives /shcr-opt-out · FHIR R4 Consent · National Data Opt-Out

The wire is standards, not a private protocol

Live

HL7 v2 and FHIR R4 over the connector layer, IHE profiles for cross-enterprise document exchange, and a canonical data model for secondary use — so a participant is not buying a dependency on commvita to talk to its neighbours.

Where it lives /connector-dashboard · /epr-hub · /ihe-profiles · /fdp-cdm
The test of a federated claim. Ask what has to be true for the architecture to work. If the answer includes “and then every organisation signs one agreement and copies its data to us”, it is a centralised system with a federated diagram. The question to put to any vendor, including this one, is simple: when a query runs, how many rows leave the organisation that holds them?

5The honest edges

The federation network is demonstrated, not deployed. The query layer, its node board and the suppression rule are built and behave as described against seeded nodes. commvita is not today operating a live multi-organisation query network, and nothing in this document should be read as saying it is.

Three entity counts here are secondary. Only the ICB figure was read from its instrument. The publishers of the other three could not be reached from this build environment, so they carry a “secondary source” tag in the figure rather than a footnote nobody reads.

Federating does not remove the need for agreements. It changes what they have to say. A partnership agreement is still signed, still has signatories and a review date, and is still the thing an information governance team is accountable for.

The Community Edition is not free. It is £1 per instance, with optional paid support, and carries the platform base — identity, access, audit, data fabric — deliberately without the clinical record or clinical workflow.

6Where every claim in this document lives

CapabilityRouteAPI / basisStatus
Resolve one person across organisations
EMPI Hub
/empiGET /empi/Live
Point at a record without copying it
NRL Connector
/nrl-connectorNational Record Locator v3.0 · FHIR DocumentReferenceDemonstrated
Shared cross-organisation record
Single Patient Record · NHR
/single-patient-record · /nhrGET /nhr/Live
Care homes contribute and read
Care Homes Shared Record
/care-homes-shared-recordPRSB Core Information StandardDemonstrated
Aggregate-only cohort feasibility, k≥5
Federated Cohort Discovery
/cohort-discoveryGDPR Art.89 · ICO anonymisationDemonstrated
Cross-site query, no rows transferred
Federated Query
/federated-queryGET /federated-query/nodes · /execute · /governanceDemonstrated
Sharing agreements and signatories
DSPA Manager
/dspa-managerGDPR Art.26 joint controllerDemonstrated
Caldicott queue, cross-controller notice
Confidential IG Spine
/ig-spineCaldicott 8 · National Data Opt-OutDemonstrated
Hash-chained access log, verified on read
Legitimate Relationship Log
/legitimate-access-logGET /legitimate-access/eventsLive
Federation governance surface
Federated Governance
/federated-governanceGET /federated-governanceLive
Person-level sharing objection
Shared Care Opt-Out
/shcr-opt-outFHIR R4 Consent · DUA Act 2025Demonstrated
Scope resolved at sign-in
RBAC and scope
/rbac-matrixeffective_scope() · effective_role()Live
Organisational map as sourced configuration
Jurisdiction Profile
/jurisdiction-profilehealth_system §7.2 · ICB Order 2026Live
Standards on the wire
PulseGrid · EPR Hub · IHE
/connector-dashboard · /epr-hub · /ihe-profilesHL7 v2 · FHIR R4 · IHE XCPD/XCALive
Canonical model for secondary use
FDP CDM Console
/fdp-cdmGET /fdp-cdm/extract/{product}Live
UK GDPR · Data Protection Act 2018Caldicott 8 PrinciplesNational Data Opt-OutFHIR R4 · HL7 v2 · IHE XCPD/XCACabinet statement, 31 July 2026 — public service boundary alignmentNon-SaMD architecture and governance surfaceCommunity Edition — £1 per instance (optional support)
© 2026 Commvita Digital Health Solutions Ltd. All rights reserved.