AlphaEngine treats the compute layer as a CoFHE processor. Keep to the
abstraction and you can swap providers without changing flows.
What the CoFHE processor does
- Encrypts at the edge: Clients encrypt amounts, directions, and constraints before leaving the browser.
- Computes on ciphertexts: Core arithmetic (add, select, compare) runs without decryption, enabling routing and batching while data stays opaque.
- Controlled reveals: Decryption is only allowed for attested operators during simulation and settlement.
- Deterministic handles: Ciphertext handles are returned to the contracts so we never store plaintext on-chain.
How it works (quick)
- Keygen: Operators provision an FHE key pair. The public key is shipped to clients; the secret key lives in the coprocessor enclave.
- Encrypt: Frontend libraries turn user inputs into ciphertexts (
ctAmount,ctMinOut,ctDirection). - Compute: The coprocessor executes arithmetic gates, producing encrypted aggregates (e.g., net order flow).
- Decrypt (permitted): Only after operator quorum and policy checks do we decrypt to simulate or settle. Contracts never see the plaintext.
Coprocessor Responsibilities
- Intent validation: Enforce bounds (minOut, deadlines, token allowlists) under encryption.
- Batch math: Sum and net orders to minimize on-chain hops and MEV surface.
- UTI simulation: Run Universal Trade Intent sims over candidate routes before posting results to the chain.
- Proof artifacts: Return receipts/attestations that EigenCompute operators can sign and post.
Trust & Access Model
- Public key distribution: Shipped in the frontend bundle and cached; rotate via config.
- Decryption permissions: Contracts tag intent handles with
allow(operatorQuorum)to limit who can open data. - Replay protection: Intents carry nonces/epochs; coprocessor rejects stale ciphertexts.
- Data retention: No plaintext persisted; ciphertext retained only for audit windows.
Performance Notes
- Latency: Expect ~sub-second client encryption; coprocessor batch compute is sized for multi-intent epochs.
- Batch sizing: Larger batches improve privacy (better netting) but add queueing delay; tune per market.
- Footguns: Mixing keys across sessions breaks decryption; ensure all clients use the latest published public key.
Inputs & Outputs
- Inputs: ciphertext handles (
ctAmount,ctMinOut,ctSlippage), market ID/pool key, expiry, user-signed metadata. - Outputs: encrypted batch nets, per-intent settlement handles, simulation receipts, optional decrypted nets (for execution only).
Observability
- Emit per-batch metrics: queue depth, batch duration, match rate, decrypt success rate.
- Log policy failures: out-of-bounds intents, expired sessions, invalid proofs.
- Surface operator IDs participating in decryption to correlate with EigenCompute attestations.