Client and patients are withheld; every figure below was measured on the running system.
Build a patient-owned health record whose breach does not become a population breach, whose consent is evidence rather than an assumption, and whose interfaces are accepted by payers and federal agencies. Centralised record stores had already shown the failure mode at national scale: one clearinghouse, roughly 190 million people exposed.
| Layer | Decision | Measured outcome |
|---|---|---|
| Data at rest | AES-256-GCM with a per-patient key derived through HKDF; ciphertext in the database | 1,047 FHIR resources and 63 patient rows encrypted in place; reads decrypt only for an authorised request |
| Key custody | Shamir secret sharing, five shares, threshold three | Any three shares reconstruct the key; two reveal nothing; shares are not co-located |
| Consent | Grant, refusal and revocation recorded as events | Revocation is provable with the file that granted access |
| Interfaces | FHIR R4 patient access with SMART scopes | Validated in CMS Blue Button and payer sandbox environments before promotion |
| Operations | Self-hosted mail, storage, search, routing and monitoring | No critical function depends on a vendor that cannot be replaced |
Read the contracts and the payer requirements; map the data flows; test what actually protects the record; then produce the decision set — build, buy or retire — in cost order. The first control or interface ships inside the first month.
Press and identity · Resume · Services · charles@trinh.io