Institutional trading where the order book cannot be inspected before execution. Orders are split into secret shares, matched blindly by multi-party computation, and settled with a proof of solvency.
A centralised dark pool still has an operator, and the operator can read the book. Encrypting it moves the problem rather than solving it, because whoever holds the key can read it too.
Shamir secret sharing removes the key. An order becomes a set of linear polynomials over a prime finite field, and any subset smaller than the threshold possesses zero information — not impractical to read, but information-theoretically nothing. Seven shares, five required: any four reveal as much as four random numbers.
Matching is the hard part, because comparing two prices usually means knowing them. Beaver triples let the nodes multiply secret shares without reconstructing anything, so cross-order volume clearing happens entirely in encrypted space. A blind equality check answers the only question that matters — is the best limit buy at or above the best limit sell — and returns one bit.
Solvency is proved rather than asserted: a trader demonstrates escrowed collateral sufficient to back an order without revealing a wallet balance or a portfolio size.
| Characteristic | Public AMM | Wall St dark pool | SMPC Dark Pool |
|---|---|---|---|
| order visibility | 100% public mempool | operator and HFT insiders | zero — cryptographically blind |
| MEV / sandwiching | extreme | asymmetric internal arbitrage | mathematically impossible |
| auditability | on-chain and transparent | a closed proprietary database | succinct zk-SNARK |
| collusion tolerance | not applicable | requires one corrupt operator | tolerates k−1 Byzantine nodes |
git clone https://github.com/KELLERBABG/SMPC-Dark-Pool.git cd SMPC-Dark-Pool && cargo test # spawn a 3-node secret-sharing committee and match two blind orders