EigenFlow: Consensus-Native Market Making
This section explains EigenFlow: how it works, why it matters, and how Aporia leverages it to build a structural execution advantage.
EigenFlow is the theoretical and operational core of Aporia's long-term competitive advantage. It is a market making framework developed specifically for Kaspa's parallel blockDAG consensus architecture — one that treats the network's transaction-ordering behavior not as an opaque black box, but as a measurable, learnable, and priceable market structure property.
This section explains EigenFlow in conceptual terms: what problem it solves, how it works, why it matters to Aporia's position in the market, and what the simulation results show. EigenFlow is a Phase 2 capability — it requires Kaspa vProg to reach its full execution potential. In Phase 1, EigenFlow-lite runs in data-collection mode, building the dataset that powers the full framework from the moment vProg launches.
Full theoretical treatment, proofs, and simulation methodology: O'Neill & O'Neill (2026).
The Problem EigenFlow Solves
Every market maker on every exchange today faces the same fundamental risk: inventory risk. When a market maker posts a bid and an offer, they are committing to buy or sell at a stated price. If the market moves against them before their quote is filled or cancelled, they absorb the loss. The wider they price their spread, the more protected they are — but the worse the experience for traders. The narrower the spread, the better the market quality — but the greater the risk.
The key variable in this tradeoff is time. Inventory risk accumulates over the period that a quote is live and exposed. A quote accepted in 50 milliseconds carries a fraction of the risk of one that lingers for 500 milliseconds. On every linear exchange — centralized or decentralized — the exposure window has a hard floor set by sequential block production: one block, one queue, one position in line.
Kaspa changes this physics. Its blockDAG produces multiple parallel blocks simultaneously. A market maker who can place quotes across multiple frontier views at once does not wait for one queue — they wait for the first of several queues. The time to the first accepted fill becomes an order statistic: statistically shorter than any single queue, shrinking further as more parallel paths are added. EigenFlow is the framework that makes this observation precise, measurable, and executable.
The core question EigenFlow answers: Given Kaspa's parallel frontier views, which paths should a market maker quote on, at what spreads, with what fees, to minimize inventory risk and maximize risk-adjusted returns? EigenFlow provides a rigorous, data-driven answer — derived from the network's own consensus behavior.
How EigenFlow Works
EigenFlow operates as a three-stage pipeline: observe the network, learn its ordering structure, and optimize market maker behavior in real time. Each stage builds on the previous, and the entire system improves continuously as more data is collected.
Stage 1 — Observe: Learning the Network's Ordering Behavior
On Kaspa's blockDAG, transaction conflicts arise naturally: when two parallel blocks contain transactions that spend the same output, only one can be accepted. GHOSTDAG resolves these conflicts deterministically, but the resolution is not random — it reflects network topology, mining pool connectivity, relay endpoint positioning, and fee incentives at the moment of the conflict.EigenFlow monitors these conflict events continuously and records which inclusion interfaces — groups of miners and relay endpoints — consistently win ordering priority. Over time, this builds an empirical picture of the network's true ordering preference structure: which paths are faster, which have stronger connectivity to the blue-chain selection, and how those properties evolve with network conditions. These inclusion interfaces are called execution classes, and the pairwise ordering preferences between them form the EigenFlow consensus kernel — a learned probability model of how Kaspa resolves competing transactions in practice.
Stage 2 — Learn: The Spectral Consensus Kernel
From the pairwise ordering data, EigenFlow constructs a Markov model of execution class competition. The stationary distribution of this model — the long-run probability that each execution class wins a conflict — is called the EigenFlow weight vector. These weights summarize, in a single compact representation, the entire observed ordering preference structure of the Kaspa network.The weights are updated on a rolling basis as new conflict events are observed, tracking changes in network topology, mining pool composition, and evolving fee dynamics. Two properties are particularly useful for risk management:
Ordering stability — the spectral gap of the model measures how stable the current ordering structure is. A large gap means ordering preferences are well-defined and predictable, supporting tighter spreads. A small gap signals a volatile or fragmented network state, warranting wider spreads and reduced sizes as a precaution.
Adversarial robustness — if a portion of the hash rate attempts to censor a market maker's transactions, the model predicts exactly how the effective acceptance rate degrades — allowing the spread formula to compensate automatically.
Stage 3 — Optimize: Real-Time Spread and Fee Computation
Armed with the EigenFlow weights, the framework computes the optimal strategy for each quoting window. For each active execution class, it determines:
Optimal spread — the bid-ask spread that maximizes risk-adjusted profit given expected acceptance time, current inventory, price volatility, and risk aversion. Because the effective exposure horizon shrinks with each additional parallel path, the optimal spread is structurally tighter than on any single-path system.
Fee schedule — miners prefer higher-fee transactions when resolving conflicts. EigenFlow computes the fee premium required to shift acceptance probability toward the preferred execution class — finding the fee level at which the marginal benefit of faster acceptance equals its marginal cost.
Staleness premium — if a frontier view is slightly outdated relative to the freshest block, EigenFlow computes a spread correction that accounts for the additional adverse selection risk from arbitrageurs who see the newer state.
In plain terms: EigenFlow watches how Kaspa resolves transaction conflicts, learns which paths win most often and why, and uses that knowledge to tell market makers exactly where to quote, at what price, and with what fee — in real time. It turns the hidden structure of blockchain consensus into a measurable, executable, and compounding market making edge.
Phase 1 vs. Phase 2: EigenFlow Integration
EigenFlow's integration with Aporia is a deliberate two-stage deployment. The distinction between phases is important to understand clearly.
Execution model
Single-path linear CLOB
Multi-frontier parallel CLOB
Order matching
Sequential, price-time priority
Parallel, acceptance-probability optimized
EigenFlow role
Data collection only (lite mode)
Full spectral kernel + live optimization
Inventory risk
Standard linear exposure window
Reduced via 1/n time concentration
MM Sharpe uplift
0% (baseline)
35–75% at 10 BPS / 60–120% at 100 BPS
Fee-aware ordering
Not applicable
Entropy-regularized, log-sum-exp HJB
AI-agent APIs
Standard REST/WebSocket
Deterministic multi-path programmable APIs
In Phase 1, EigenFlow operates in lite mode: it observes the Kaspa network, records conflict-resolution events, and builds the training dataset for the spectral consensus kernel — but it does not influence execution. The matching engine remains a single-path linear CLOB. The EigenFlow data pipeline runs in parallel, silently accumulating the ordering preference data that will power Phase 2.
This design is intentional. The spectral kernel requires a meaningful dataset of conflict observations to produce reliable EigenFlow weights. By running the data pipeline from Aporia's launch, the full model will be trained and validated before Phase 2 execution begins. When Kaspa vProg launches, Aporia will activate multi-frontier quoting with an already-calibrated model — not a cold-start system.
The data advantage: Every day Aporia operates in Phase 1 is another day of conflict-resolution data being collected that no competitor can replicate without also building on Kaspa. The EigenFlow kernel is not just software — it is a trained model whose quality depends on proprietary network observation data accumulated over time. This is Aporia's deepest and most durable competitive moat.
Why EigenFlow Matters for Aporia
EigenFlow is not merely a technical feature — it is the mechanism through which Aporia's structural advantage translates into durable business outcomes across four compounding dimensions.
Tighter Spreads by Design
On a linear exchange, spread width is determined by volatility, risk aversion, and exposure time. EigenFlow structurally reduces the third factor through parallel frontier quoting. The result: market makers on Aporia can profitably quote tighter spreads than on any competing linear venue. This is not an incentive program; it is a consequence of execution physics. Tighter spreads mean better prices for every trader on the platform.
Attracting Professional Market Makers
Professional market makers compete on Sharpe ratio — risk-adjusted returns — not just absolute profit. A 35–75% Sharpe improvement is not a marginal edge; it is the difference between a viable and a non-viable strategy for many quantitative trading firms. EigenFlow makes Aporia the structurally preferred venue for any market maker benchmarking on a risk-adjusted basis. As EigenFlow data improves and Phase 2 activates, this preference deepens: the model becomes more accurate, optimal spreads tighter, and the advantage over linear venues larger.
A Self-Reinforcing Data and Liquidity Flywheel
EigenFlow creates a flywheel that is difficult to replicate even if competitors copy the framework design: more volume generates more conflict observations; better observations produce more accurate weights; more accurate weights enable tighter spreads; tighter spreads attract more market makers; more market makers produce deeper liquidity; deeper liquidity attracts more volume. The flywheel is self-reinforcing and data-intensive — both properties that reward the first mover.
Lower Execution Risk
Parallel quoting across frontier views concentrates time-to-fill as an order statistic
→
Tighter Spreads
Reduced inventory risk enables MMs to quote narrower bid-ask spreads profitably
→
Better MM Economics
Higher Sharpe ratios attract top-tier professional market makers to Aporia
Deeper Liquidity
More MMs quoting tighter means deeper books and higher fill rates for traders
→
More Volume
Better prices and deeper books attract retail and institutional flow
→
Better Consensus Data
Greater activity generates richer conflict data, sharpening the EigenFlow kernel
The EigenFlow Liquidity Flywheel — each stage reinforces the next, compounding with data
Built for AI Agents
As autonomous trading agents move on-chain, they require deterministic, programmable, verifiable execution with transparent ordering. EigenFlow exposes acceptance probabilities, execution class weights, and optimal spread parameters as structured, queryable outputs. An AI agent building a liquidity provision strategy on Aporia consumes the EigenFlow kernel's output directly — without having to model ordering behavior from scratch. This makes Aporia the natural infrastructure layer for autonomous market making as AI agent trading scales on Kaspa.
Simulation Results
The EigenFlow research paper reports Monte Carlo simulation results across 10,000 trials using Kaspa post-Crescendo parameters (10 BPS, GHOSTDAG k = 124). The simulation measures net-of-fee market maker Sharpe ratio improvement versus a single-path baseline, across n = 1, 2, and 3 active parallel frontier views.
Simulation parameters:
Trials: 10,000 per configuration
Horizon: H = 1,000 blocks (~100 seconds of trading)
Per-block volatility: σ = 0.002 (~20% annualised), consistent with Kaspa market conditions
Fill intensity: A = 1.0, κ = 1.5 (exponential fill model)
Risk aversion: γ = 0.1 (representative professional quantitative MM)
Propagation floor: D = 0.5 blocks (= 50ms), accounting for network latency
1
Single-path baseline, 10 BPS
0.0773
Baseline
2
2 parallel paths, 10 BPS
0.1022
+36%
3
3 parallel paths, 10 BPS + vProg
0.1202
+63%
–
Overall simulation range, 10 BPS
–
35–75%
–
Projected range, 100 BPS upgrade
–
60–120%
EigenFlow simulation results — net-of-fee Sharpe ratio vs. single-path baseline. Source: O'Neill & O'Neill (2026), https://zenodo.org/records/18379248
The results confirm the theoretical prediction: each additional parallel path reduces the standard deviation of time-to-accepted execution and improves net Sharpe ratio. The improvement from n = 1 to n = 2 is +36%; to n = 3 is +63%. Both figures are net of the fee premiums required to achieve ordering priority on additional paths.
The 35–75% simulation range reflects sensitivity to network conditions — specifically, the spectral gap. In stable states with a large spectral gap, the improvement is closer to 75%. In more volatile states, closer to 35%, as the model widens spreads defensively. In all conditions, EigenFlow outperforms the single-path baseline.
At Kaspa's projected 100 BPS upgrade: As Kaspa scales toward 100 blocks per second, the expected tip count increases, creating more parallel frontier views per quoting window. Modelled Sharpe improvements reach 60–120% under multi-path quoting with stable propagation latency — a further structural doubling of the advantage. The EigenFlow framework scales directly with Kaspa's throughput upgrades without requiring protocol changes on Aporia's side.
EigenBatch: The Phase 3 Mechanism Design Extension
Looking beyond Phase 2, the EigenFlow research paper describes EigenBatch: a vProg-native micro-batch auction mechanism that extends the framework to collective execution across multiple market participants simultaneously. Rather than each market maker optimizing quote placement independently, EigenBatch aggregates competing quotes into a single clearing event proven by the vProg execution environment.
The key property of EigenBatch is the structural elimination of stale-tip routing attacks: because execution is determined by a single cryptographically proven clearing rule rather than by individual tip selection, adverse selection risk from outdated frontier views is removed at the mechanism level rather than managed through spread widening. This represents a further step-change in market quality beyond what Phase 2 parallel quoting achieves.
EigenBatch is a Phase 3 R&D target. Full equilibrium analysis is ongoing. For preliminary design, see: https://zenodo.org/records/18379248
Last updated