Why
The four capabilities look unrelated and are the same move four times. Clearing, matching, routing and resolution each name a property the on-chain design already has for free — and each is the price of keeping that property once the system has to be fast, capital-efficient, or connected to the world. That symmetry is the finding, and it is what stops the list from being a syllabus.
Netting is the clearest case. Full collateralisation is not a safety feature the chain gives you; it is what remains after the clearing house is removed. Novation, multilateral netting, margin models and the default waterfall exist to make capital go further, and every scheme that recovers that efficiency — portfolio margin at the account level, a hybrid engine with an off-chain credit pool — re-introduces exactly the credit relationship whose absence was the selling point. The honest framing is not how do we net on chain but what are we willing to trust again, and how much capital does that buy back.
Matching repeats it with a different noun. The chain is the audit trail only while the order lifecycle happens on it; move matching off-chain for cost and latency and the trail leaves with it. Sequencer log commitments, Merkle proofs and ZK state-transition proofs are how a hybrid venue buys back auditability it stopped getting for free — and the residual question is always which parts of the lifecycle remain unprovable, since amendments and cancellations are harder to commit to honestly than fills.
Routing and resolution close the pattern. There is no best-execution duty on chain, but the economics of routing return as MEV — an agency cost paid to whoever sequences, which auctions, private order flow and slashable solver commitments try to price rather than abolish. And the chain cannot see the world at all, so every oracle fallback, dispute escrow and TLS-attested feed is a trusted party added back one at a time, while surveillance — the invariant checks, circuit breakers and rate limits that a traditional venue staffs a desk for — has to be written as code that runs without anyone watching.
Which is the positioning argument, stated precisely. The scarce engineer is not the one who can deploy a token or an AMM. It is the one who can read a nineteenth-century clearing mechanism, identify which of its guarantees a chain already provides and which it merely removed, and then re-implement the second group under gas, asynchrony and permissionlessness. event-contract-plumbing established that the requirements list is public and free to read. This card is what reading it costs.
How it works
The four rows, and the number each one is measured by
| Capability | What the chain gives free | What re-acquiring it costs | The number |
|---|---|---|---|
| Clearing and netting — novation, multilateral netting, SPAN/VaR margin, default waterfall; on-chain portfolio margin and hybrid credit pools | No counterparty risk, because every position is fully collateralised | A credit relationship comes back. Netting without a CCP means somebody is exposed between the trade and the cover | Required collateral ÷ net exposure. 1.0 means no netting exists; the gap below 1.0 is what was bought and from whom |
| Hybrid matching — off-chain LOB with batch settlement; sequencer log commitments, Merkle proofs, ZK state-transition proofs | The chain is the audit trail, publicly and by construction | The trail leaves with the matching. It has to be re-bought cryptographically, and amendments and cancellations are the hard part, not fills | Share of the order lifecycle provable to an outsider — submissions, amendments, cancellations, fills |
| Intent routing — Dutch and batch auctions, private order flow, solver networks with slashable commitments | No best-execution obligation to satisfy | The obligation returns as MEV, paid to whoever sequences. Auctions price the agency cost rather than removing it | Realised slippage against the executable side of the book — ask when buying, bid when selling, never the mid |
| Oracle failure and surveillance — multi-oracle fallback, dispute escrow, TLS-attested feeds, MPC feeds; invariant checks, circuit breakers, rate limits | Deterministic settlement, once the input is known | The chain cannot see the world. Every fallback adds a trusted party, and surveillance must run with nobody watching | Time to resolution when the primary oracle is unavailable, and how many parties had to be trusted to get there |
Why the symmetry matters more than the list
Read the middle column downward and the same sentence appears four times: the free property was free because something had been removed, and putting the capability back puts the thing back too. That is not an argument against building any of it — a venue that refuses all four is slow, capital-hungry, unauditable and blind. It is an argument for stating the price each time, because a design that adds netting without naming the credit it re-introduced has not solved counterparty risk; it has stopped measuring it.
The order to learn them in, and why
Not the order above. Start with resolution, because a market that cannot settle correctly makes every other capability irrelevant, and because it is the row with the least public prior art. Then netting, which decides the capital structure everything else runs on. Then matching, which is a performance decision that can be deferred until volume exists. Routing last, because MEV is a cost that only appears at size — and by then the first three have determined how much of it there is to extract.
The honest limit of a card like this
Four capabilities is a career, not a sprint, and a list of things to learn is the least reliable artifact in this catalogue — it survives contact with reality only if each row is reduced to a number that can be produced for a real system. That is why the fourth column exists, and why a row without one should be dropped rather than admired.
Review clarification
Why use a chain if trust comes back?
The remaining value is not no trust but constrained trust. A sequencer may order trades without receiving custody or power to rewrite final settlement; an oracle may decide an event without authority over unrelated treasury funds. The deepest test is credible exit: if one service disappears, do users retain their assets, history and ability to continue elsewhere? If not, the chain may be extra cost around an effectively centralised system.
Every efficiency number needs a risk counter-number
| Capability | Efficiency measure | Risk counter-measure |
|---|---|---|
| Netting | 1 - net exposure / gross exposure |
collateral / stressed potential loss |
| Matching | lifecycle steps with proofs | meaningful deviations still undetectable |
| Routing | slippage from executable bid/ask | benchmark age, impact, accessible venues |
| Oracle fallback | time to resolution | incorrect-resolution risk and source independence |
Collateral below net exposure may be under-collateralisation, not netting. A high provable fraction can hide censorship. Slippage mixes routing quality with latency and impact. Fast oracle recovery can mean one unsafe fallback. The evaluation must pair efficiency gained with risk reintroduced.
Where it lands in Jayverse
- Verex: name the credit relationship before adding netting. If portfolio margin or a hybrid credit pool is ever added to the CLOB, track required collateral ÷ net exposure as an explicit number — 1.0 means no netting exists, and the gap below it is what was bought and from whom.
- Verex: if matching ever moves off the CLOB, buy back the audit trail on purpose. Sequencer log commitments or Merkle proofs of the order lifecycle should cover amendments and cancellations, not just fills, and the provable share of the lifecycle should be a published number.
- OFA: price the routing obligation, don't remove it. The solver auction should measure realized slippage against the executable side of the book (ask when buying, bid when selling), and slashable solver commitments are what makes the MEV cost explicit rather than hidden.
- Auditor / Devnet: the oracle-failure row is a surveillance job, not a feature. Write the invariant checks, circuit breakers and rate limits as code that runs without anyone watching, and track time-to-resolution when the primary oracle is unavailable as a standing number.
Key expressions
| Expression | 뜻 · 쓰이는 자리 |
|---|---|
| buy back | 되사다, 내준 것을 다시 사들이다 · 체인이 없앤 속성을 값을 치르고 되찾는 비유. "each buying back what the chain gave away" |
| novation | 채무 갱개(경신) · 청산 과정에서 계약 당사자를 교체하는 법률 용어. "Novation, multilateral netting, margin models" |
| default waterfall | 디폴트 손실 분담 순서 · 청산소가 손실을 정해진 순서로 흡수하는 체계. "margin models and the default waterfall" |
| credit relationship | 신용(여신) 관계 · 누군가에게 상환 위험을 지는 관계. "re-introduces exactly the credit relationship" |
| best-execution duty | 최선집행 의무 · 중개인이 고객에게 가장 유리한 조건으로 체결해야 할 법적 의무. "There is no best-execution duty on chain" |
| price the agency cost | 대리인 비용을 값으로 매기다 · 위임에서 생기는 손실을 가격에 반영하는 것. "auctions price the agency cost" |
| credible exit | 신뢰할 수 있는 탈출구 · 서비스가 사라져도 사용자가 자산을 지킬 수 있는지 보는 시험. "The deepest test is credible exit" |
| syllabus | (여기서는) 공부 목록, 강의계획표 · 실행할 가격표가 아니라 읽을거리 목록이 되는 것을 경계. "rather than a syllabus" |
| under-collateralisation | 담보 부족 상태 · 순노출 대비 담보가 모자란 상태. "may be under-collateralisation, not netting" |
| staff a desk for | 그 업무를 위해 전담 데스크(인력)를 두다 · 전통 거래소가 감시 인력을 배치하는 것을 가리킴. "a traditional venue staffs a desk for" |
| CCP | 중앙청산소(Central Counterparty) · 청산소가 매도자·매수자 사이에서 신용위험을 흡수하는 전통 금융 기관. "Netting without a CCP means somebody is exposed" |
| LOB | 지정가 주문서(Limit Order Book) · 오프체인에서 매칭이 이루어지는 전통적 주문 장부 구조. "off-chain LOB with batch settlement" |
| MEV | 최대 추출 가능 가치(Maximal Extractable Value) · 체인 상 거래 순서 결정 권한에서 나오는 대리인 비용. "the economics of routing return as MEV" |
| TLS | 전송계층 보안(Transport Layer Security) · 오라클 피드가 신뢰를 부여받는 방식 중 하나, 암호화 통신 증명. "a TLS-attested feed is a trusted party" |
| MPC | 다자간 계산(Multi-Party Computation) · 여러 참여자가 비밀을 나누어 계산하는 오라클 피드 방식. "dispute escrow, TLS-attested feeds, MPC feeds" |
| SPAN | 표준 포트폴리오 분석(Standard Portfolio Analysis of Risk) · 전통 청산소가 쓰는 마진 계산 방법론. "novation, multilateral netting, SPAN/VaR margin" |
| VaR | 위험가치(Value at Risk) · 일정 확률로 발생 가능한 최대 손실을 추정하는 위험 지표. "SPAN/VaR margin, default waterfall" |
| ZK | 영지식(Zero-Knowledge) · 데이터를 공개하지 않고 유효성만 증명하는 암호 기법, 여기선 상태 전이 증명. "Merkle proofs and ZK state-transition proofs" |