Decentralized Clinical Data AI Training
DeSci meets clinical AI infrastructure
Decentralized science (DeSci) applies Web3 primitives — token incentives, smart contracts, proof verification, and community governance — to scientific research workflows. In clinical AI, the core DeSci value proposition is: verifiable, incentive-aligned access to clinical data for AI training without centralised gatekeepers.
Rapha Protocol implements this through a specific architecture: compute-to-data routing, hardware attestation, Polygon mainnet proof anchors, Node NFT ownership, and USDC clearing vault settlement. The protocol does not tokenise patient data. It tokenises compute access rights and settlement flows.
Polygon mainnet settlement — 70/20/5/5
Every training job flows through RaphaClearingVault on Polygon mainnet (Chain ID 137):
- 70% hospital treasury — the institution hosting the data and edge appliance.
- 20% Node NFT owner — the entity that owns the RaphaNodeNFT representing the physical edge appliance.
- 5% node promoter/onboarder — the entity that facilitated the hospital deployment.
- 5% Rapha Protocol treasury — protocol development, operations, and security.
Settlement is proof-gated: the clearing vault only releases funds after verifying a trusted attestor signature on a proof digest that encodes the job ID, node token ID, AI company wallet, escrow amount, hardware hash, and payload hash. Replay protection prevents double settlement.
Node NFT ownership economy
RaphaNodeNFT (0x19432C08a4806f961D0ec589c6B68fe258E34d07) is an ERC-721 token representing yield rights to one physical Rapha edge appliance. Each NFT is bound to an immutable hardware hash (TPM/SGX identity). The current owner receives 20% of every USDC settlement flowing through that node.
This creates a market-based infrastructure model: node ownership rewards deployment of physical edge compute capacity. The more nodes deployed in hospitals, the more compute-to-data capacity exists for AI researchers. The more jobs processed, the more revenue flows to node owners — creating a flywheel that aligns infrastructure investment with AI research demand.
Proof receipts and auditability
Every training job produces a public proof receipt anchored on Polygon mainnet. The receipt commits to:
- Job ID and node token ID.
- AI company wallet address.
- Processed record count (from RaphaDataLoader).
- Output hash (SHA-256 of trained model weights).
- Zero-raw-PHI-export flag (true for all production jobs).
- SGX/DCAP evidence and TPM evidence.
- OPA policy verdicts.
The receipt proves execution, not clinical validity. It is a cryptographic commitment, not a regulatory certification. It provides auditable evidence that a specific model was trained against a specific dataset at a specific time — without exposing the underlying clinical data.