CareEZ operates a governance inversion: the clinical dataset and derived models are co-stewarded by HKU Swallowing Research Laboratory and Full Linkage Community Care Association (s.88 charity) — neither of which is Carewells Limited.
Impact money in Hong Kong is rightly cautious about subsidising a private moat. The CareEZ two-layer design removes that concern by separating the grant-funded public asset from the commercial operations of Carewells Limited at the constitutional level — not just at the level of accounting policy.
Grant funds three things and only these three things: (i) a public clinical-data resource on Hong Kong dysphagia and 照護食 adoption; (ii) the city's first Hong Kong 照護食 Adoption Index; and (iii) the RCHE Adoption Acceleration Programme across 30 care homes. The data, the Index methodology, the trained models, and the carer-facing knowledge hub produced under the grant are co-governed by HKU Swallowing Research Laboratory and Full Linkage Community Care Association — a Section 88 charity that is a separate legal entity from Carewells. Carewells does not own this asset, cannot privatise it, and cannot license it for exclusive commercial benefit.
Carewells Limited's commercial operations — enterprise screening deployment for institutional buyers, 照護食 certification and compliance tooling sold to private operators, B2B institutional contracts, and downstream nutrition services — run on a separate corporate track. The commercial layer never operates on grant-funded data and never inherits grant-funded IP. The public asset is published under the co-governance protocol; Carewells builds its commercial tools against the same public reference, exactly as any other vendor in Hong Kong may.
The grant funds the public-good adoption layer. The commercial layer sustains operations and is the post-grant engine. These two tracks do not cross.
The co-steward structure is the data-governance answer to the most material objection an HKCSS or research-ethics reviewer can raise: who owns the clinical dataset, and what stops the applicant operator from privatising a public resource? Two independent institutions hold that answer.
Anchors the clinical-research integrity of the public dataset. Owns classifier validation thresholds, voice-biomarker reference standards, and Adoption Index methodology. The laboratory's standing in Hong Kong's clinical community makes data-quality standards externally auditable and immune to commercial pressure from Carewells.
A registered Hong Kong s.88 charity — a separate legal entity with its own board, governance, and audited accounts. Full Linkage's s.88 status anchors the public-interest disposition of the data: its constitutional objects prohibit private benefit, making it structurally impossible for the dataset to be redirected to commercial ends without breaching charity law.
All software infrastructure developed under the grant-funded public layer is released under the Apache License 2.0. This is not a defensive PR choice — it is a structural requirement of the public-asset commitment.
The CareEZ carer-facing AI assistant, the IDDSI classification engine, the 照護食 Adoption Index
methodology codebase, and all grant-funded training materials are published under
Apache-2.0. Any operator, regulator, or researcher in Hong Kong — or globally — may
inspect, fork, audit, or deploy the same infrastructure without requiring a Carewells licence or
commercial agreement.
The commercial layer (enterprise audit-log module, certification tooling, B2B compliance dashboards) is separate Carewells IP and is not subject to this licence. The line between public and commercial layers maps exactly to the two-layer architecture described above.
This means: HKCSS, the Hospital Authority, any NGO operator, or any future public custodian can take the public-layer codebase and run it independently. Carewells is not a chokepoint.
Clinical AI for dysphagia screening is public-health infrastructure. If it is built as proprietary software by a single vendor, the sector becomes dependent on that vendor's pricing, continuity, and priorities. Apache-2.0 removes that dependency: the code is auditable by any clinical reviewer, forkable by any public agency, and deployable by any qualified operator. The governance design makes CareEZ's contribution durable beyond any single company's lifespan.
The broader clinical knowledge base at dysphagia.cn (600+ structured IDDSI and 照護食 reference articles in four languages) is similarly published as a free, open reference — no paywall, no login required.
After the grant closes (Month 24), the public asset continues under its co-governance protocol with HKU and Full Linkage. Its recurring operational cost — model inference, hosting, knowledge-hub maintenance, governance secretariat — totals under HK$300,000/year on conservative projections.
That recurring envelope is covered, in full, by commercial-layer revenue from Year 2: powder-line margin alone clears the public-asset hosting and inference bill; certification fees and NGO licensing form the buffer. No further public funding is required to keep the asset alive. The commercial layer is the post-grant engine; the public asset is the legacy.
The grant funds the public-good adoption layer. Once the 24-month grant closes, the public asset continues under its co-governance protocol; the commercial layer continues under Carewells' own P&L; neither requires further public funding. This is structural sustainability, not aspirational.
繁體版本籌備中 — Traditional Chinese version in preparation. For governance enquiries: [email protected].