zkFMI
日本語

Comparison with other systems

Eighteen products and platforms whose customers have problems that overlap with this stack, read from their own specifications, announcements and regulatory filings on 2026-09-05. The result is not a scorecard. It is a list of the axes on which a difference can be stated, the axes on which somebody else is ahead, and the sentences this project may and may not say as a consequence.

source: zkfmi-crypto/docs/research/COMPETITORS_EX_CANTON_2026-09-05.md · CANTON_NETWORK_2026-09-05.md · CANTON_FABLE_5_1_MAX_REVIEW_2026-09-05.md · COMPETITIVE_DECISIONS_2026-09-05.md (v1.2) · docs/strategy/ZKFMI_STRATEGY_2026-09-05.md · 54 distinct primary sources for the survey, 37 of 38 URLs re-opened for the Canton review

Three rules were applied throughout and should be applied when reading. Nothing was measured: no competitor's environment was run, no code audited, no trade executed, so fees, throughput, latency and funding are never ranked. A commercial announcement is a fact reported by that company, not an operational audit. And "unverified" means not found in the materials read; it is never scored as "unsupported".

How to read the stage labels

labelmeaning here
commercial case announcedan operator or adopter says the target workflow ran. Not an independent operational audit.
mainnet / alpha announceda network stage. Distinct from commercial delivery in a regulated market.
adoption / demonstration / plannedkept at the stage the announcement states. A planned date passing does not make it live.
specification confirmedthe operating model was checked in documentation. A particular deployment's status is unverified.
unverifiednot confirmed in the materials read. Different from unimplemented or unsupported.

zkFMI's own baseline, read the same day: research software on one host, with smoke evidence for two fills settled in one transaction and for two continuous trading rounds. Asset identifiers and settlement metadata on the native note path are public. Post-quantum hybrids were standalone at that date and have since passed a single-host integration acceptance; see Post-quantum migration. Aethel is excluded from the eighteen because it is an application on the stack, not a competitor.

All eighteen against zkFMI

One card per target, Canton included, and zkFMI last under the same four questions, the ones the comparison found decisive. Green means the property holds in the documented standard path, amber that it holds in part or depends on the application, grey that it does not or is out of scope. Under each card is the one sentence that distinguishes zkFMI from that target. It is a difference, not a ranking: on stage, every target is ahead of zkFMI.

holds in the documented standard pathin part, or depends on the applicationdoes not, or out of scope

Canton (in depth)

mainnet 2024, commercial repo platform, live Treasury trade, three Japanese PoCs
operator does not see the order
no, own validator and application read plaintext views
rules verifiable by an outsider
relevant parties re-execute Daml; no portable proof in the standard path
reservation → atomic settlement
yes, Token Standard allocations, one transaction, in plaintext
settlement hides amount and price
from unrelated parties and synchronisers only

difference from zkFMI Canton limits who reads the plaintext; zkFMI withholds the complete order from every computing node under a 2-of-7 bound and lets an outsider verify the rule from a proof. Canton is ahead on everything else, and could host a zkPI verifier.

Renegade

mainnet, Arbitrum One
operator does not see the order
the delegated relayer reads them; a self-run relayer avoids it
rules verifiable by an outsider
matching correctness is proved; the order set and admission order are not
reservation → atomic settlement
match settled on-chain; no advance reservation described
settlement hides amount and price
confidential wallet state

difference from zkFMI Midpoint crossing at an external price versus price-time priority or RFQ over a finalised admission order; a relayer boundary versus participant-side splitting; no eligibility or reservation leg.

Prime Match

live at J.P. Morgan, 2023 paper
operator does not see the order
a semi-honest bank hub between two clients
rules verifiable by an outsider
no
reservation → atomic settlement
bank pipeline
settlement hides amount and price
bank sees it

difference from zkFMI Two-party inventory matching with the bank in the middle and no settlement leg; zkFMI is seven-node, provable and settled. Prime Match is the prior that refutes any "first" claim.

Penumbra

specification confirmed; sealed-bid batches are a future feature
operator does not see the order
swap inputs and amounts are public in the current specification
rules verifiable by an outsider
chain-verified batch execution
reservation → atomic settlement
no reservation concept
settlement hides amount and price
transfers and claims yes, swaps no

difference from zkFMI A per-block batch auction with public swap inputs versus a continuous secret book or RFQ. Unlinkable wallets and secret prices are different rows and Penumbra has the first, not the second.

Dusk

mainnet specifications, NPEX collaboration
operator does not see the order
model dependent; the exchange path uses public accounts
rules verifiable by an outsider
chain validation
reservation → atomic settlement
DvP designs described
settlement hides amount and price
confidential transfers optional

difference from zkFMI A confidential ledger for regulated assets without confidential market computation; zkFMI adds the venue and the proof of best execution. Neither has a licensed clearing house behind its clearing code.

Arcium

mainnet alpha
operator does not see the order
MPC, confidentiality with one honest party
rules verifiable by an outsider
application defined
reservation → atomic settlement
application defined
settlement hides amount and price
application defined

difference from zkFMI A general MPC platform, dishonest-majority with abort; zkFMI is one specific workflow, admission to settlement, on honest-majority MPC. The same workflow could be built on Arcium, which is the alternative to evaluate.

Zama

mainnet 2025-12, a sealed-bid auction 2026-01
operator does not see the order
FHE; decryption by a threshold KMS, t < n/3
rules verifiable by an outsider
input proofs of knowledge and coprocessor signatures, not a proof of the business result
reservation → atomic settlement
application defined
settlement hides amount and price
application defined

difference from zkFMI Encrypted-state computation with a separate decryption authority versus MPC with proofs; a platform versus a workflow. Cost and completion behaviour under one workload are unmeasured on both sides.

Aztec

alpha V5; proving-system vulnerability disclosed 2026-08
operator does not see the order
client-side proving, the witness never leaves the client
rules verifiable by an outsider
for the client's own transaction
reservation → atomic settlement
application defined
settlement hides amount and price
private state

difference from zkFMI One party proving over its own secrets versus several firms computing over each other's; zkFMI's matching needs the second. Neither is audited.

Hyperledger Fabric

specification confirmed, widely deployed
operator does not see the order
authorised peers read the private data
rules verifiable by an outsider
no
reservation → atomic settlement
chaincode
settlement hides amount and price
hidden from other organisations only

difference from zkFMI Access control decides which organisations see the data; zkFMI is for the case where the computing organisation itself must not. Where that case is absent, Fabric is the simpler answer.

R3 Corda

adopted for CSD Prague settlement
operator does not see the order
participants see; a non-validating notary sees state references
rules verifiable by an outsider
participants validate, no outside proof
reservation → atomic settlement
contract state and uniqueness
settlement hides amount and price
from non-participants

difference from zkFMI Limited sharing among the parties to a contract, no confidential computation and no proof of a market rule; a large adoption record. An integration candidate for the zkPI.

Kinexys

commercial, JPM Coin on Base 2026-04
operator does not see the order
the bank
rules verifiable by an outsider
no
reservation → atomic settlement
bank-run workflows
settlement hides amount and price
the bank

difference from zkFMI A bank's own payment and tokenisation service with customers; zkFMI has no cash leg and would need one like it.

Fnality

sterling system operating under Bank of England limits
operator does not see the order
the system operator
rules verifiable by an outsider
no
reservation → atomic settlement
payment arrangements
settlement hides amount and price
operator

difference from zkFMI Central-bank-backed digital cash; the cash leg zkFMI lacks. No secret price formation is intended.

Partior

live payments
operator does not see the order
participating banks
rules verifiable by an outsider
no
reservation → atomic settlement
PvP; a DvP proof of concept
settlement hides amount and price
banks

difference from zkFMI Bank-network PvP with plaintext among members; zkFMI's PvP passes a scalar by adaptor signature between ledgers that share nothing. Candidate cash leg.

Ownera / FinP2P

APIs and orchestration plans published
operator does not see the order
each institution's router
rules verifiable by an outsider
signatures and receipts
reservation → atomic settlement
intents, holds, orchestration per ledger capability
settlement hides amount and price
routers

difference from zkFMI The closest to where zkPI sits: it coordinates instructions across ledgers in plaintext with signatures. zkPI binds a proof of hidden propositions to the instruction. "Connecting ledgers" is not the difference.

Swift shared ledger

ready for initial use 2026-07, pilots preparing
operator does not see the order
Swift and the banks
rules verifiable by an outsider
no
reservation → atomic settlement
commitments between banks, final settlement in existing systems
settlement hides amount and price
banks

difference from zkFMI Interbank coordination of tokenised deposits at Swift's scale; nothing hidden from the banks, nothing computed. Monitor.

Chainlink

features described 2026-05
operator does not see the order
plaintext inside a TEE
rules verifiable by an outsider
oracle-network signatures
reservation → atomic settlement
cross-chain messages
settlement hides amount and price
private tokens described

difference from zkFMI Trust placed in hardware enclaves and a signing network versus MPC and proofs; the word "verifiable" covers different things. A candidate carrier for a zkPI.

Progmat

commercial ST issuance; Avalanche L1 migration announced; JGB repo PoC on Canton
operator does not see the order
issuer, trust and distributor
rules verifiable by an outsider
no
reservation → atomic settlement
ST practice on the platform
settlement hides amount and price
the roles

difference from zkFMI Japanese security-token issuance and administration in production; zkFMI would be an add-on for eligibility and collateral computation hidden from those roles. Priority Japanese comparison.

BOOSTRY / ibet for Fin

operating since 2021
operator does not see the order
consortium members
rules verifiable by an outsider
no
reservation → atomic settlement
standard contracts
settlement hides amount and price
members

difference from zkFMI A member-run Japanese ST network with standardised contracts; the same add-on relation as Progmat.

zkFMI

research, one host, one administrator
operator does not see the order
no node holds a complete order, at most two of seven malicious; traffic observable
rules verifiable by an outsider
zkPI verifier and quote proof, for the stated order set, not the whole market
reservation → atomic settlement
maker and taker reservations, batch DvP, smoke evidence
settlement hides amount and price
amount and price hidden; asset tag and metadata public on the native rail

difference from zkFMI The composition: secret from the computing nodes, rule proved to an outsider, reservation carried unopened into settlement. No production, no independent operators, no audit.

Read the first two questions across the cards and the shape of the field appears. The institutional rows are grey on both because they are built for a world where the operator is trusted; the infrastructure rows are green on the first and amber on the second because they leave the workflow to the application; the trading rows split. zkFMI is green on both, and grey on stage, which is the whole position in one line.

Nine decisions

The comparison closes on nine numbered decisions. They were reviewed a second time, at maximum effort, with Canton added; all nine were retained and several gained evidence.

#decisionwhyconsequence
C01MPC, ZK and a proof that fills were computed correctly do not make zkFMI uniqueRenegade's VALID MATCH proves input validity and matching correctness. Canton has the relevant participants re-execute Daml rules. Prime Match ran financial MPC in 2023.never market "only we can prove market rules"
C02The promising axes are who learns which secret and which rules and states are bound to the same transactionRenegade's delegated relayer reads the wallets it serves. Canton's host validator reads all of its parties' data. OCLOB splits on the participant's side, so no single node holds an order, under a 2-of-7 collusion bound.first establish which disclosure a customer accepts; target the projects that need secrecy from the computing entities themselves
C03Price formation differs, and that is not superiorityRenegade documents midpoint crossing at an external price. QOMM is RFQ over confidential pricing rules; OCLOB is price-time priority over a finalised admission order.no "speed ranking" that mixes RFQ, continuous books and midpoint crossing
C04zkFMI is not uniformly stronger in secret-sharing securityArcium's Cerberus keeps confidentiality with one honest party and aborts on deviation. Zama's KMS completes key operations with t < n/3. OCLOB assumes at most two of seven malicious.state confidentiality, correctness, completion and operator independence as four separate claims
C05Instruction standards, DvP and cross-ledger coordination already existOwnera describes intents, holds and orchestration. Corda verifies contract state and uniqueness. Canton's Token Standard (CIP-0056) standardises allocations and one-transaction settlement of all legs.propose zkPI as verification that can be added to existing contracts. Atomicity itself is not the claim.
C06zkFMI is a research MVP with real function and a maturity gapTwo fills in one transaction and two continuous rounds are smoke evidence on one host. Canton has a mainnet, a commercial repo platform, a multi-firm pilot, a live Treasury trade and launched PoCs.next priorities are continued use, consistency under failure and independent operation. Do not understate the other side's maturity.
C07In Japan, connecting to existing securities workflows and to a cash leg may be a condition of adoptionProgmat, ibet for Fin, Kinexys and Fnality publish adoption records. MUFG and Progmat started a JGB repo PoC on Canton in August 2026; JSCC, Mizuho and Nomura a JGB collateral PoC in April.the first adoption demonstration is an integration, not a replacement
C08System-wide quantum resistance is not yet a competitive advantageStandalone hybrids and now the merged main branches of all eight repositories do not cover commitments, proofs, stored state or every transport. Competitors' PQC status is unverified, not absent.offer PQC as a migration plan per boundary
C09Canton is at once the priority competitor and a candidate platform or integration destinationDaml holds business logic, authorisation and disclosure; synchronisers can be global or private. Wiring external MPC results or a zkPI verifier into a Daml app is a design inference, not a verified implementation.keep the standalone DeFMI L1, QOMM, OCLOB and DeCCP objectives; evaluate a Canton integration at the second gate

The shorthand used elsewhere on this site, "others prove circuits, we prove markets", is a description of what the quote proof states. It is not, on its own, evidence of novelty, and the comparison retired it as such.

Canton, the closest comparison

Canton is not one product. It is a protocol, a set of synchronisers, validators that host parties, and Daml applications written by third parties. Reading any Canton case as evidence for all Canton capabilities was the most common error found in the earlier drafts, so the layers come first.

layerwhat official materials confirmkeep separate from
Canton Networkindependently operated applications, validators and several synchronisers connected together. Each validator stores only the ledger fragments of the parties it hosts.one firm's Canton project as evidence that the whole network's capabilities are live
Daml applicationcontracts define business logic, signatory/observer/controller roles, authorisation and disclosure. They execute on validators.properties of the foundation versus order, credit, reservation and clearing rules that each application implements
validator (participant node)hosts parties, stores their contracts, decrypts and validates the transaction views addressed to themthe public-chain sense of "validator" in which every validator reads every transaction
synchroniserthe sequencer orders and delivers encrypted messages; the mediator collects confirmations from the relevant validators and declares commit or reject. Neither validates Daml contents.ordering and confirmation versus validation of business logic and input correctness
Global Synchronizera public synchroniser run by Super Validators with CometBFT ordering, tolerance f = ⌊(n−1)/3⌋, traffic paid by burning Canton Coin, sponsorship required to join. Applications may instead use private synchronisers.the assumption that every Canton application runs through it
application providersupplies on-ledger Daml and off-ledger logic, authentication and submissionprotocol guarantees versus trust that the provider admits, selects and submits inputs honestly

Who reads what

Canton splits a transaction into views and encrypts each for the participants of its informees. Unrelated parties receive nothing and synchronisers cannot decrypt. That is sub-transaction privacy: plaintext shared only with the parties that need it. It is a different guarantee from the one this stack aims at, which is that no computing node holds a complete order within the tolerated collusion bound.

entitycan readcannot read
party (signatory, observer, controller)views it is an informee of; divulgence may expose contracts it is not a stakeholder inpayload and party metadata of unrelated views
the party's own validatorthe trust model states it "sees all your data and could block you". External party keys keep signing authority with the party but are not described as removing visibility.
counterpartyplaintext of the shared views, trusted not to leak them
application provider, off-ledger componentsimplementation dependent. An off-ledger matching engine may handle plaintext orders.
sequencerencrypted envelopes, recipients, order, size, timingcontents
mediatoreach view's informee list, confirmation policy, each approve or rejectcontents
unrelated parties and validatorsnothing
auditorplaintext of the relevant views if designed in as an observera portable proof of a proposition without receiving the secrets is not part of the documented standard path

This is a design choice with a stated reason, not a gap in documentation. The 2020 Canton whitepaper says advanced cryptography including MPC is computationally expensive and that Canton limits visibility under stronger trust assumptions instead; a 2025 official post calls zero-knowledge proofs for general privacy experimental. Nothing prohibits a Canton application from calling external MPC or verifying an external proof, so the difference is between the documented standard path and this stack's path, and cannot be written as "impossible on Canton".

Axis by axis

axisCanton, confirmed scopezkFMI, current positionjudgment
who learns the secrethidden from unrelated parties and synchronisers; host validator and entitled parties read the relevant viewsparticipant-side splitting, no MPC node holds a complete input; collusion of three or more nodes and traffic observation remain limitsa difference can be shown and it shrinks to nothing if the customer is content to disclose to its own validator
computation hidden from the computing entitiesthe standard path decrypts views and re-executes Daml in plaintextseven-node MPC, at most two malicious, research implementationcandidate external MPC on Canton is a design option with no verified implementation; superiority undetermined until compared under one workflow, collusion bound, availability and cost
eligibility, credit, reservationsDaml can express and verify all three; reservation-based settlement is already standardised by CIP-0056 allocations, in plaintextDeKYX eligibility, maker and taker reservations, guarantee capacity across the end-to-end pathnot a feature difference the difference is whether the order-to-reservation correspondence is hidden from the matching nodes or read by registry, app and validator
market rulesparties validate deterministic Daml logic; an identical CLOB or RFQ application is unverified, the one documented trading app is an AMM that submits swaps off-chainQOMM RFQ and OCLOB price-time priority computed in MPC, bound into the zkPIbuildable on Canton in principle; not shown to exist
input setbinds the contracts a submitted transaction references; completeness of off-ledger admission is at the application boundaryaims to bind admission order and order set into the proof; pre-admission censorship remains possibleneither proves "the whole market"; compare admission commitments and target sets
settlement atomicitycross-application DvP in one Daml transaction on a common synchroniser; across synchronisers, reassignment then atomic settlement, with assets pending in betweensmoke evidence for multi-fill batch settlement on the native railno zkFMI claim until compared with the same legs including external ledgers
third-party verificationrelevant validators check their views; an auditor can be an observera zkPI verifier checks defined propositions without the secretspotential value not exclusive, since a zkPI verifier could sit inside a Daml application
operational trustown or outsourced validator, counterparties, application provider, synchroniser, Foundation governanceMPC nodes, coordinator, read-back service, DeFMI validators; independence of keys and operators not acceptedcompare who is responsible for confidentiality, correctness, availability, censorship and recovery; not node counts
maturity, performance, legal framemainnet since 2024-07, protocol upgrades live, commercial repo platform, one live Treasury trade, three Japanese PoCs launchedsingle-host research MVP, smoke evidencedo not understate Canton performance under equal conditions, demand and authorisation are unverified on both sides
post-quantumunverifiedhybrid signatures and KEM on the merged main branches of all eight repositories; ZK relations classicalmark unverified, claim neither direction

Cases, at the stage their sources state

casestage in primary sourcesdo not generalise to
Global Synchronizergo-live 2024-07-01; Canton 3.5 logical synchroniser upgrade live 2026-06-29usage, SLOs or authorisation of any individual application
Canton Network Pilot2024-03-12: 22 dApps, more than 350 simulated transactions on TestNet; 45 participants in the summary, 36 in the role-by-role bodyproduction assets or continuing commercial trading
Broadridge DLRmigrated to Canton in 2023 with cash off-chain (2024 customer story); self-reported average daily repo of $280bn in August 2025Global Synchronizer use, on-chain cash, a confidential order book
Tradeweb2026-07-01: one transaction, tokenised US Treasuries against USDCx, synchronised on-chain settlementsustained capacity or a full lifecycle
DTCCplan to mint a subset of DTC-custodied Treasuries on Canton; MVP planned for a controlled production environment in the first half of 2026, completion unverifiedcommercial launch
JSCC / Mizuho / Nomura2026-04-20: PoC for digital JGB collateral management, keeping JGBs' legal status under the book-entry transfer law; selected for the FSA's Payment Innovation Projectcompletion, end date, commercialisation
MUFG / Progmat / Secured Finance2026-08-13: PoC to bring JGB repo on-chain, account management institutions updating their transfer books alongside the chain, tokenised deposits or stablecoins as the cash legcompletion or legal delivery
Progmat / DCC working grouplaunched 2026-05, report targeted for 2026-10a platform choice

The Japanese PoCs matter for this project specifically. They sit in the workflows the application survey ranks highest, collateral and repo, and they keep the JGB inside the book-entry system exactly as the regulation page says a Japanese deployment must. The strategy therefore does not propose replacing those workflows. It looks for the parts inside them that need an input hidden even from the computing entity: the order, the rate quote, the collateral allocation.

Group by group

The notes behind the table, for the seventeen other than Canton. Priority is a judgment of relevance to a build, buy or connect decision, not a market-share ranking. "Integration candidate" means a plausible destination for a zkPI, not an existing integration; none exists.

Institutional ledgers, cash legs and connectivity

targetcompeting taskconfidentiality and trust boundarystagetreatment
R3 Cordaasset state and inter-institution workflowsharing among relevant parties; a non-validating notary sees input-state references, not full contentsadopted for CSD Prague's DLT settlement; a separate "Corda protocol" for Solana was planned for H1 2026 and is not the ledger productpriority comparison, integration candidate
Kinexys (J.P. Morgan)bank payments, tokenisation, fund flowsbank-operated; distinct from hiding inputs from the bankJPM Coin on Base and a first fund-flow transaction announced 2026-04-28institutional comparison, cash-leg candidate
Fnalityinstitutional digital cashbacked by central-bank account funds; secret price formation not its focussterling system in the Bank of England's FMI report, operating under limitscash-leg candidate
Partiorcross-border payments, FX PvPnetwork of participating banks; confidential market computation unverifiedlive payment cases; a DvP proof of conceptcash-leg and PvP candidate
Ownera / FinP2Passet distribution and settlement instructions across ledgersrouters per institution, adapters, agreement among participants; depends on the underlying ledgers' hold and atomic capabilitiesAPIs and orchestration plans publishedpriority comparison, the reference for integration design; closest to where zkPI sits
Swift shared ledgerinterbank payment coordination, tokenised depositsa shared layer run by Swift, bank ledgers on the side, final settlement through existing systemsready for initial use 2026-07-09, banks preparing live pilotsmonitor
Chainlink CCIP, CRE, ACEexternal data, policy and cross-chain executionoracle networks, TEEs and distributed key generation; plaintext inside a TEE is a different trust placement from MPC or ZK, whatever the word "verifiable" is applied toprivacy features and cases described 2026-05integration candidate and comparison for zkPI
ProgmatJapanese security-token issuance and administrationissuer, trust and distributor roles; hiding inputs from all of them unverifiedST project list; provider announcement of a completed migration from Corda to a dedicated Avalanche L1; a tokenised investment-trust demonstration without external sales, chain unstatedpriority Japanese comparison, integration candidate
BOOSTRY / ibet for FinJapanese STs, consortium operationmember-run network, standard contractsoperation launched 2021-06; body text of the consortium page could not be extractedpriority Japanese comparison, integration candidate

The common lesson from this group: connectivity and customer relationships, not cryptography, are the barrier. A zkPI adds something only if the receiving ledger verifies the proof; a receiver that accepts a signature or a digest has turned the guarantee back into trust in the signer. The six problems every instruction source has to solve on the applications page is the list of what a destination must be able to do.

Confidential orders and confidential asset trading

targetmarket and verificationconfidentiality and trust boundarystagetreatment
Renegadepairwise MPC produces a collaborative SNARK (VALID MATCH: valid inputs, correct matching, encrypted outputs); midpoint crossing at an external price; balances updated on-chainthe relayer a customer delegates to reads that customer's orders and balances in plaintext; other relayers do not; a customer may run its own relayermainnet on Arbitrum Onepriority comparison for OCLOB. The difference is the delegation boundary, price-time priority over a finalised admission order versus midpoint crossing, and continuity from fill to reservation to settlement. Whether Renegade proves that no eligible order was omitted was not audited and is not asserted.
Prime Match (J.P. Morgan)MPC inventory matching between a bank and its clientsbank-and-client threat model set by the authors; not this stack's node topologylive operation reported in the 2023 paper; 2026 status unverifiedprior work; refutes any "first financial MPC" claim. Full treatment on the prior-art page.
Penumbrashielded pool, per-block batch swapsswap input assets and amounts are public in the current specification; sealed-bid batches are a stated future featurespecification confirmedmonitor market design. Wallet unlinkability and pre-execution price secrecy are separate rows and must not be merged.
Duskledger for regulated assets with confidential transfers and selective disclosuremodel dependent; the exchange integration guide specifies a public account modelmainnet specifications; collaboration with NPEXmonitor securities workflow. A partner's licence does not extend to applications on the chain; the same standard applies to DeCCP, whose code is not a licensed CCP.

Confidential computation and application infrastructure

targetprovidesconfidentiality and trust boundarystagetreatment
Arciumgeneral-purpose MPC execution environments on SolanaCerberus: dishonest majority, confidentiality with one honest party, detect-and-abort; availability is a separate conditionmainnet alphapriority comparison and a procurement option. "We have seven nodes, they are centralised" does not hold; building the same workflow on Arcium is an alternative to evaluate.
ZamaFHE over encrypted state, confidential tokens, threshold KMScomputation, decryption authority and key management are separate; the KMS assumes a strong honest majority t < n/3; the coprocessor separates input proofs of knowledge, FHE evaluation, commitments and signaturesmainnet 2025-12-31, a sealed-bid auction in 2026-01infrastructure comparison alongside Arcium. Do not read its input proofs as a proof of a whole business result, and do not claim only zkFMI has external verification.
Aztecprivate and public applications on an Ethereum L2 with client-side provingprivate witnesses stay on the client; not joint computation over several firms' inputsAlpha V5; a critical proving-system vulnerability disclosed 2026-08-07 with a fix planned for V6, completion unverifiedmonitor. This stack has no independent audit either and gains nothing from a competitor's alpha status.
Hyperledger Fabricpermissioned ledger, private data collectionsactual data to authorised peers, hashes to the channel; orderers receive no private data. Limits which organisations read; does not hide inputs from the authorised peersspecification confirmedthe in-house alternative. Where secrecy from everyone is not required, existing governance plus Fabric may be the clearer choice, and the case for zkFMI must be made from the boundary access control cannot reach.

The same standard applied to this stack

The comparison read this project's own documents with the same rules, and fixed the results by SHA-256 in a baseline file so that later drift cannot improve them retroactively.

itemjudgmentevidence and limits
separation of orders from the MPC nodesdesign and implementation of the participant-side path confirmedseven nodes, coordinator never sees an order; the browser-compatible demo still holds plaintext
settlement of several fillsconfirmed from existing artifactstwo fills, one transaction, fourteen confirmations, roots agree across five validators and a restart; not rerun for the comparison
continuous tradingsmoke onlytwo rounds and three fills; cancellation, expiry, reuse of returned assets, restart of seven MPC nodes; the earlier rejected run is kept in the ledger
ledger finalitytrust in the read service remainseach node reads back but does not verify an independent consensus proof
settlement confidentialitynot every fieldasset identifiers and settlement metadata on the native rail are public
identity and eligibilitypseudonymous within a scopeDeKYX is not issuer-unlinkable and is not a certified KYC product
clearingresearch state machineDeCCP is neither a custodian nor a licensed clearing house
cryptographic securityfixed defects distinct from unaccepted guaranteesa note-proof defect was found and fixed; the Triptych dependency is experimental; no independent audit
QOMM economicsconfirmation experiment not passedsynthetic-data smoke; no claim of price improvement or customer benefit
independent operationfalseone host, one administrator, no WAN

Wording this project uses, and wording it refuses

sentencestatuscondition, or what to say instead
"a research implementation that connects confidential order admission, price-time priority matching, advance reservation and proof-carrying settlement"usedwith the path and the single-host scope stated
"two fills settled atomically in one transaction"usedwith the run's conditions and artifact; never as capacity
"aims to let an outsider verify that a stated order set, the rules and the settlement agree"conditionalnaming the propositions the verifier checks and the residual trust in signers and read-back
"the first financial MPC" / "the only MPC-plus-ZK settlement"refusedPrime Match and Renegade precede it
"competitors prove circuits, we prove market rules", as evidence of noveltyrefusedRenegade proves matching; Canton's participants validate Daml rules. Name the concrete function, set, authorisation and settlement that differ.
"Canton cannot do orders, eligibility or reservations" / "Canton has no atomic DvP"refusedboth exist; compare documented individual applications
"seven nodes, therefore stronger than Arcium"refusedthe honesty assumptions differ
"a confidential ledger hides everything, including asset types, traffic and identities"refusedasset tags on the native rail are public; traffic is observable; DeKYX is linkable by its issuer
"production FMI or CCP" / "the whole system is quantum-resistant" / "faster than competitors"not yetneeds institutional evidence, evidence covering every cryptographic path, and measurements under equal conditions respectively

What zkFMI must show before claiming a difference

These are verification tasks for this project, derived from the comparison. They are not a list of anyone else's weaknesses.

candidate propositionevidence requiredmeasured against
orders and pricing rules hidden even from the operatorthe list of entities with visibility from participant-side share generation through MPC to settlement; collusion conditions; every plaintext point; key administratorsRenegade, Arcium, Prime Match, Canton
execution rules verifiable afterwardspropositions binding the order set, admission order, eligibility, price-time priority and partial fills, with an independent verification procedureRenegade, Penumbra, market apps on Zama, Daml apps on Canton
computation result equals deliverythe same commitment bound through computation, instruction, reservation and consumption of both legs; state unchanged by retry or double spendOwnera, Corda, Chainlink, Canton Token Standard
adoptable by an institutionissuance, custody, registration and cash-leg responsibilities; recovery; supervisory disclosure; entry and exit; legal finalityKinexys, Fnality, Partior, Swift, Progmat, ibet for Fin, Dusk, Canton
long-term confidentiality and verifiabilitya threat model separating signatures, KEM, commitments, proofs, TLS and stored data, with migration, revocation and re-verification pathsevery adopted configuration

What the comparison changed

The strategy that follows from the nine decisions is short. The first use case is institutional secondary trading of digital securities where an order must stay hidden from the computing entities, meaning the organisation's own validator, the application operator and the matching engine, until its order in the book is fixed. The first thing offered is an integration module: confidential matching, advance reservation and a zkPI verifier added to a ledger the customer already runs, with the standalone DeFMI L1 kept as the reference path. Canton and Renegade are the comparisons for markets and settlement, Arcium and Zama for confidential computation, Ownera for integration design.

Before any of that, the first gate is a customer question, not a build: does at least one organisation document that it needs inputs hidden from its own validator and matching engine? If the answer everywhere is that disclosure to one's own validator is acceptable, a Canton-style design suffices for those customers and the initial use case is wrong. The comparison names that as the condition under which the strategy is revised, and it has not been tested.

When these conclusions change

  • acceptance receipts for continuous trading, cancellation, expiry, contention, recovery and independent operation (C02, C06);
  • an explicit mapping of the same admission set, ordering, eligibility, reservations and settlement onto Canton or Renegade (C01, C03, C09);
  • the same end-to-end path run on Arcium or Zama with measured differences in output, confidentiality scope, cost and failure behaviour (C04);
  • a destination ledger confirming APIs, hold, commit, abort and read-back and the allocation of legal responsibilities (C05, C07);
  • migration receipts covering components beyond signatures and key exchange (C08);
  • any change in a vendor's version, stage or fix notice, which requires rereading the source before reuse.

What this comparison did not do

It did not run a Canton node, use a competitor's application, benchmark anything, audit anyone's code, interview a customer or write to any external system. Some sources could not be read in full: the Renegade whitepaper returned a server error, the Canton pilot PDF exceeded the retrieval limit and was replaced by the official announcement of the same content, the Tradeweb investor page timed out and was replaced by Canton's republication, and the ibet for Fin consortium page yielded only its search summary. Each substitution is recorded in the source documents, and nothing unread is treated as read. The comparison is dated; a stage label from 2026-09-05 describes that day.

Primary sources checked (selection, all opened 2026-09-05)