Background & Foundation
This section provides background on the evolution of decentralized exchanges, the limitations of existing approaches, and the technical foundation of Kaspa's BlockDAG consensus.
The Evolution of Decentralized Exchanges
Decentralized exchange design has advanced through four distinct generations, each solving a real problem while leaving a deeper structural limitation unaddressed. Understanding this lineage is essential to appreciating why Aporia represents a genuine architectural discontinuity rather than an incremental improvement.
Generation 1 — Automated Market Makers (AMMs).
Uniswap and its successors solved the cold-start problem for decentralized liquidity. By replacing the order book with an on-chain pricing formula (x·y = k and its variants), AMMs enabled any token pair to attract liquidity without a professional market maker. The tradeoff was substantial: AMMs price by formula rather than by order discovery. Large trades face significant slippage. Liquidity providers bear impermanent loss. Capital efficiency is poor because liquidity is spread across the entire price curve rather than concentrated where it is needed. AMMs democratized access to liquidity; they did not democratize access to professional trading.
Generation 2 — Off-chain Order Books with On-chain Settlement.
Protocols such as dYdX v3 recognized that on-chain order book maintenance was computationally infeasible at competitive latency and cost. Their solution was to run a traditional centralized matching engine off-chain and settle net trades on-chain. This recovered performance but reintroduced a trust problem: the off-chain operator could censor orders, front-run trades, or selectively favor certain participants. The on-chain settlement layer could verify that two orders crossed; it could not verify that price-time priority was respected, or that other orders were not suppressed.
Generation 3 — Application-Specific L1 Order Books.
Hyperliquid took a more radical approach: build a purpose-built Layer 1 blockchain (HyperCore) optimized for financial workloads, with a consensus algorithm (HyperBFT, inspired by HotStuff) fast enough to support a fully on-chain central limit order book (CLOB). The result is 200,000 orders per second with one-block finality, all on-chain and transparent. This is a genuine achievement. Yet it remains fundamentally linear: HyperBFT produces one block at a time in a single ordered sequence. Every order, cancel, and trade exists in one canonical queue. The execution physics are identical to a centralized exchange — only the custody model has changed.
Generation 4 — zk-Proven Order Book Rollups.
Lighter introduced SNARK-based proofs for complete financial operations, including price-time-priority order matching. Built as an application-specific Layer 2 on Ethereum, Lighter achieves cryptographically verifiable fairness: a market participant can independently verify that their order was processed according to the rules, without trusting the operator. It inherits Ethereum's security and censorship-resistance for exits. It is the most rigorous approach to trustless order book trading available today. And yet: Lighter's sequencer processes transactions in a single ordered queue. The execution model is linear. The structural inventory risk that market makers face on every other exchange is present here too.
The evolution of decentralized exchange design, summarized:
AMM DEX
Uniswap v2/v3
Trustless liquidity pools
Slippage, impermanent loss, no price discovery
Off-chain OB + settlement
dYdX v3
CEX-grade matching speed
Operator trust, censorship risk
App-specific L1 CLOB
Hyperliquid
Fully on-chain order book at scale
Single-path linear execution, MEV exposure
zk-Rollup CLOB
Lighter
SNARK-proven price-time priority
Single sequencer, linear execution
Parallel-native CLOB
Aporia
Consensus-native execution physics
Requires BlockDAG infrastructure (Kaspa)
Each generation solved a problem in the trust layer. None has yet addressed the physics layer.
The Structural Limitation of Linear Execution
The shared constraint across every exchange ever built — centralized or decentralized, first generation or fourth — is linear transaction ordering. In every existing system, transactions form a total order: transaction A is processed strictly before transaction B, which is processed strictly before transaction C. This total order is what makes bookkeeping tractable, settlement deterministic, and auditability possible. It is also what creates the inventory risk problem.
For a market maker, the relevant quantity is not the spread they post — it is the spread they must post to remain profitable. That spread is determined by the cost of bearing inventory risk, which is proportional to the expected time their quote remains live and exposed to adverse price moves. On a linear system, this exposure window has a hard floor: the time between when a quote is submitted and when it achieves finality in the canonical ordering. No amount of throughput improvement eliminates this window. A system processing one million transactions per second still produces a queue; the queue still has a front and a back; a quote at the back is still exposed for the duration of its wait.
Formal framing: In the classical Avellaneda–Stoikov market making model (2008), the optimal bid-ask spread is proportional to γ · σ² · T, where γ is risk aversion, σ² is price variance, and T is the exposure horizon. On a linear system, T is bounded below by block time and finality. The only lever available is spread width. EigenFlow replaces the fixed T with a DAG-extended exposure horizon that shrinks as parallel paths are added.
There is a second, related limitation: ordering opacity. In centralized exchanges, transaction ordering is a policy decision made by the matching engine operator, invisible to participants. The operator can internalize the advantage — routing certain order flow to preferred counterparties, for example — without any external verifiability. In decentralized systems with a single sequencer, the sequencer operator holds equivalent power: it can selectively favor certain maker orders, delay others, or manipulate matching outcomes within the constraints of what the on-chain settlement layer checks. This is the root of MEV (maximal extractable value): economic rent captured by whoever controls transaction ordering.
Aporia addresses both problems simultaneously. The parallel structure of Kaspa's consensus reduces the exposure window structurally. The EigenFlow framework makes ordering probabilities observable and priceable, turning a hidden policy into a transparent, competitive market parameter.
Kaspa and the GHOSTDAG Protocol
Kaspa is a proof-of-work blockchain that generalizes Nakamoto consensus to a directed acyclic graph structure. Rather than requiring miners to extend a single canonical chain — discarding all parallel blocks as orphans — Kaspa's GHOSTDAG protocol incorporates parallel blocks into a unified DAG, assigns them a blue-set membership based on accumulated proof-of-work, and derives a deterministic total order over all transactions from the DAG structure.
The key parameters and their implications for Aporia are as follows:
Block rate (post-Crescendo)
10 blocks per second (BPS)
10x more execution paths than 1 BPS chains
GHOSTDAG parameter k
k = 124 (KIP-14)
Defines the blue-set; governs parallel tip count
Expected tip count
2–3 simultaneous tips
2–3 parallel frontier views per quoting window
Conflict resolution
Transaction-level mutual exclusivity
Only one fill per quote pair; no double-fill risk
Security threshold
½(1 − ε) honest hash power
Strong adversarial robustness for MM strategies
Roadmap block rate
~100 BPS (future upgrade)
Up to 10x further execution-time concentration
Two properties of GHOSTDAG are especially important for Aporia's design. First, transaction-level conflict resolution: when two parallel blocks contain conflicting transactions (such as two fills of the same quote), GHOSTDAG's ordering determines which transaction is accepted. Both blocks remain valid; only one transaction is executed. This mutual exclusivity is the property that makes multi-path quoting safe — a market maker cannot be double-filled by placing quotes on parallel paths. Second, the blue-set selection: GHOSTDAG identifies a "blue" subDAG representing the work contributed by honest miners. The ordering within the blue-set is stable and predictable, making the spectral analysis that underpins EigenFlow tractable.
Why Kaspa and not other blockDAGs? Kaspa is the only proof-of-work blockDAG operating at production scale with a stable, well-specified conflict resolution protocol (GHOSTDAG / DAGKnight). Its post-Crescendo 10 BPS rate is 10× higher than Bitcoin's block rate, producing 2–3 simultaneous tips per quoting window. Its forthcoming 100 BPS upgrade will extend this further. No other live network offers the combination of parallel block production, transaction-level mutual exclusivity, and the security guarantees of proof-of-work consensus that Aporia requires.
Market Making Theory: From Classical to DAG-Extended
The academic foundation for EigenFlow draws on the stochastic optimal control literature for market making, extending it to the branching execution environment of a BlockDAG. A brief review of the classical framework clarifies what is new.
Classical market making (Avellaneda–Stoikov, 2008)
The seminal model treats a market maker who posts bid and ask quotes around a mid-price that follows arithmetic Brownian motion. The market maker's problem is to choose spread widths that maximize expected utility (under CARA preferences), accounting for inventory risk: the risk that accumulated inventory loses value before it can be unwound. The optimal solution is a Hamilton–Jacobi–Bellman (HJB) equation whose solution yields spread formulas proportional to the product of risk aversion, price variance, and the remaining time horizon. In this framework, reducing the effective time horizon T is the only way to tighten spreads without reducing risk aversion or price volatility — neither of which a market maker controls.
The DAG extension
On a BlockDAG, the market maker's optimization problem changes fundamentally. Instead of a single execution path with a fixed time horizon, there are n competing frontier views, each with its own inclusion clock. The EigenFlow framework (O'Neill & O'Neill, 2026) derives a DAG-extended HJB equation that incorporates the full distribution of transaction acceptance across parallel paths. The key result is that the effective exposure horizon τ is no longer fixed — it is a function of how many paths are active and how the network's ordering behavior (the EigenFlow spectral kernel) weights those paths. As more paths are added, τ shrinks. Spreads tighten. Market maker profitability improves. All of this is derived from first principles, without assuming any coordination mechanism or policy intervention.
Fee-aware ordering
A further extension endogenizes the role of transaction fees in determining acceptance probability. Miners are modeled as preferring higher-fee transactions when choosing among conflicting options. EigenFlow incorporates this via an entropy-regularized control formulation: the market maker can pay fees to shift acceptance probabilities toward preferred paths, and the optimal fee schedule is derived as part of the HJB solution. The result is that ordering advantage — historically opaque and policy-driven — becomes a transparent, priced, and competitive market parameter on Aporia.
The full theoretical treatment, including proofs of all core theorems, simulation methodology, and adversarial robustness analysis, is available in the EigenFlow research paper:
O'Neill, A. & O'Neill, R. (2026). EIGENFLOW: Optimal Market Making on Directed Acyclic Graph Blockchains.
Last updated