Frequently asked questions

No. There is no token, no sale, and nothing here is an investment. The repositories are MIT-licensed research code. Bonds that nodes post and that misbehaviour slashes are a deployment's decision and are denominated in whatever the deployment chooses; the code carries relative weights only.

The settlement layer runs as a dedicated non-EVM Avalanche L1: AvalancheGo hosts consensus and launches a Rust virtual machine over the RPCChainVM protocol. The venues (QOMM, OCLOB) are not on any chain; they are seven-node multiparty computations that emit instructions. DeKYX, DeCCP and Aethel are deterministic state machines a host embeds. So "a blockchain" describes one component, and the design would settle on any ledger that can verify the instruction natively. See Architecture.

The existing non-PQC path uses Pedersen commitments, range/product/equality proofs and Triptych-based note membership. It verifies conservation, non-negative remainders, quantity times price and non-reuse. Public outputs, timing and other metadata remain visible. New PQC research uses hash/code-based collaborative proofs, but operational notes and the full venue path remain incomplete. See DeFMI and PQC.

The code is published under github.com/zkFMI. It has run end to end on one Linux host: five validators, seven MPC parties, seven order-book containers. It has not run across independent organisations, on a WAN, or with real assets. See Status.

Because the verifier's arithmetic does not fit. One ed25519 scalar multiplication in EVM bytecode measured 302,401 gas. Solana has ristretto255 syscalls but no scalar-field syscall, and a one-of-many verifier's polynomial evaluation exceeds the compute ceiling at a crowd of 64. A Subnet-EVM precompile was feasible but would make the curve code a consensus bug on a chain nobody else watches. A native Rust VM keeps the verifier in the audited crates. See the execution-layer comparison.

Because a TEE adds the CPU manufacturer's attestation key, the enclave implementation and microarchitectural side channels to the trust boundary, and the premise of the venue is that no party is trusted with the order flow. The threat model says so explicitly against Tesseract-style designs. See Security.

Seven nodes with threshold T=2 give an honest-majority Shamir setting where malicious security costs about 1.2× the wall clock. Three colluding nodes reconstruct every secret; that is stated, not hidden. Seven is also where robust reconstruction falls one short: at n ≥ 4t+1, nine nodes, a decoder alone gives guaranteed output delivery, and that is built and measured. The cost of two more nodes is two more institutions. See Accountability.

A sigma response z = k + c·w is affine in the witness, so seven nodes holding shares of w each compute their piece and the pieces Lagrange-combine into a proof an ordinary verifier accepts, while no node holds w. A general-purpose SNARK has no such structure; collaborative SNARKs run the whole prover inside MPC. The price is that the MPC must run over the commitment group's scalar field, which costs 2.00× traffic and 1.07× wall clock. See Binding.

The MPC is 70 rounds deep and each round is a network round trip, so a quote is 0.6 s in one metro and 26 s with nodes on two continents. Whether it matters depends on whether the price moved: measured on UniswapX, 26 s of drift is the same size as the dispersion the market already has inside one block, on the harshest rail available. See Choosing a deployment.

Distinguish hybrid signatures/key exchange, the existing classical note proofs, and fresh-network PQC proof research. Hash/code-based collaborative proofs have reached bounded ledger transitions in recorded research runs. Complete operational PQC note relations, venue integration and security of the exact post-quantum composition remain incomplete. See current PQC work.

Three grains of selective disclosure are built: one committed field opened without the others; one scope of a wallet (an instrument, a quarter, a mandate) handed to a named auditor through a view key that reads that scope's incoming notes and cannot spend; and a reconciliation statement a quorum assembles about a total none of them holds. A grant is signed and names the grantee, so a key found where it should not be traces to the grant. What is not built is a real supervisor's key ceremony and legal-request workflow. See DeFMI and Regulation.

Partly, and the cryptography does not touch this because a firm price is what the protocol exists to return. Measured, the correlation between probes and inventory is about 0.53 and does not grow with the probe budget; what grows is the confidence. A per-entity cap therefore sets how often a picture can be refreshed, not whether one is obtained. Publication of winning quotes is under a differential-privacy budget drawn inside the MPC. See Deployment and QOMM.

Effective anonymity depends on candidate selection, other users' activity, timing, venue and reservation metadata. The historical ring experiment measured an observer under a particular synthetic traffic model; it does not guarantee anonymity for the current implementation. The asset cohort size alone is not an effective anonymity-set guarantee. See public information and limits.

The note path verifies candidate sets, nullifiers, output commitments and ciphertexts. The current Triptych-based proof does not give the sender the spending secret or publish the public owner key as the nullifier. Senders still know what they sent, and venue, reservation and other metadata can link activity. See how nullifiers work.

Non-PQC notes do. A note is an asset-management unit; its Pedersen value commitment is a separate layer. Conservation uses additive homomorphism and proofs, alongside ownership and non-reuse checks. See the worked explanation.

Commitments, recipient ciphertexts, nullifiers and proofs are public. Amounts and blindings, decryption keys and spending keys remain private. Copying a public note does not grant spending authority, but outputs of one transaction and timing remain visible. A public cryptographic format does not imply transaction unlinkability. See the public/private boundary.

Self-custody of one's own assets differs from a business managing keys that can transfer client assets. With MPC shares, the combined powers of related businesses still matter. Requirements also depend on asset classification and the custodian's regulated role. See primary sources and design questions.

Those match secret orders; some do so in MPC, and that is not claimed as new. The differences OCLOB tests are continuous rather than batched processing, variable quantity with partial and multiple fills, arrival order fixed 5-of-7 before content is opened, a per-level aggregate public book, and settlement that needs no signature after the match. QOMM's difference is that what is audited is the market mechanism (best execution) rather than circuit correctness, and that the same unopened commitment is carried into settlement. See Prior art.

Prime Match is in production and faster; it trusts a semi-honest bank at the hub and its outcome is not third-party auditable. QOMM has no semi-honest party, produces an auditable best-execution proof, and is not faster on any reading. Both halves are stated. See Prior art.

DvP on DLT assumes the price is public. Here the price is a commitment the quote proof shows is minimal, the quorum signs, the instruction carries and the ledger's product proof consumes, never opened. The netting, overdraft-limit, waterfall and novation constructions are the standard FMI ones with commitments in place of numbers. See DeFMI.

No. Simultaneous exchange of two parties' tokens is not novation. DeCCP is the component that supplies novation, netting, margin and a waterfall, and it is a state machine, not a licensed clearing house. Under Japan's book-entry regime DeFMI cannot be the register; it mirrors one and reconciles. Both are on the may-not-claim list. See Regulation.

Yes. Any system that decides a transfer can issue a typed instruction: a fund administrator, a collateral engine, a bank's sanctions check, a registrar. The recommended first product is fund subscription and redemption. See Use cases.

cargo test --locked --workspace in any repository, then the zkPI vector check. The five-validator Avalanche gate needs Linux and an explicitly supplied AvalancheGo. See Get started.

A settlement authority no single party can open, operators that are independent in fact and not only in arithmetic, persistent state generations in authenticated storage, real authentication and key management, a proof of the private computation that validators can check, and legal finality. Each is listed with what exists today. See Status and Roadmap.

Because a claim about what a system does only means something next to a claim about what it does not, and because most of the design decisions came from predictions that missed. See How this project works.