Entity Toxicity Scoring

01Executive Summary

JIL Sovereign's execution router evaluates every trading entity across five dimensions of toxicity to detect and mitigate adverse selection in real-time. The scoring model measures trader edge relative to mid-market price, historical win rate, directional flow bias, inventory skew impact, and LP spread capture rate. Each dimension contributes to a composite toxicity score that determines the entity's enforcement tier.

Graduated enforcement ensures proportional response: low-toxicity entities experience wider spreads, medium-toxicity entities face reduced position sizes, high-toxicity entities are restricted to RFQ-only with deposit requirements, and critical-toxicity entities are blocked entirely. Critically, the system includes Sybil cluster detection that identifies correlated wallets and treats them as a single entity for scoring purposes, preventing adversaries from fragmenting toxic flow across multiple addresses.

02Problem Statement

Liquidity providers on decentralized exchanges face systematic losses from informed traders who exploit information asymmetry. These traders consistently trade ahead of price movements, extracting value from LPs through adverse selection. The problem is compounded by Sybil attacks where a single entity operates dozens of wallets to circumvent per-wallet protections.

2.1 Adverse Selection Attack Patterns

• Informed Trading: Entities with access to faster information sources (CEX prices, mempool data, oracle updates) consistently trade before price movements, causing LPs to fill at stale prices.
• One-Way Flow: Toxic entities exhibit strong directional bias - they predominantly buy before upward moves and sell before downward moves, extracting maximum LP value.
• Sybil Fragmentation: To avoid detection thresholds, toxic entities split their activity across multiple wallets, each appearing benign individually but collectively generating significant toxic flow.
• Inventory Manipulation: Coordinated trading across multiple pairs to skew LP inventory in unfavorable directions, amplifying LP losses across the portfolio.

2.2 Why Existing Approaches Fail

03Technical Architecture

The toxicity scoring system operates in three layers: per-wallet metric collection, Sybil cluster aggregation, and entity-level enforcement. All five scoring dimensions are computed at the entity level after cluster aggregation.

3.1 Five Scoring Dimensions

3.2 Enforcement Tiers

3.3 Sybil Cluster Detection

The cluster detection algorithm identifies correlated wallets using four signals: temporal correlation (wallets that submit orders within narrow time windows), directional correlation (wallets that consistently trade in the same direction), funding chain analysis (wallets funded from common sources), and behavioral fingerprinting (wallets exhibiting identical order size patterns). Wallets identified as belonging to the same cluster are merged into a single entity, and their aggregate activity is scored as one.

04Implementation

4.1 Scoring Pipeline

1. Trade ingestion: Every completed fill is recorded with the entity ID (post-cluster aggregation), direction, size, fill price, mid-market price at fill time, and timestamp.
2. Window calculation: Scoring metrics are computed over a rolling window (default 7 days) using exponential decay weighting so that recent activity has more influence than older trades.
3. Dimension scoring: Each of the five dimensions is scored independently on a 0 to 100 scale based on how far the entity's behavior deviates from the baseline (random trading) distribution.
4. Composite score: The weighted sum of all five dimension scores produces the composite toxicity score (0 to 100).
5. Tier assignment: The composite score determines the enforcement tier, which is cached and applied to all subsequent orders from any wallet in the entity's cluster.

4.2 Cluster Aggregation

When a new wallet is detected as part of an existing cluster, its historical trades are retroactively attributed to the cluster entity. The entity's toxicity score is recalculated with the merged data, which may cause an immediate tier change. This retroactive attribution prevents the "clean wallet" evasion where an adversary creates a new wallet to reset their score while continuing the same toxic strategy.

4.3 Appeal Mechanism

Entities flagged at HIGH or CRITICAL tiers can submit an appeal through the dispute resolution service. Appeals are reviewed against the scoring evidence, and if the entity can demonstrate that the scoring was based on legitimate market-making activity rather than adverse selection, the score can be adjusted manually by governance.

05Integration with JIL Ecosystem

5.1 Execution Router

The execution router checks the entity's toxicity tier before routing any trade intent. HIGH toxicity entities are forced to the RFQ lane with deposit requirements. CRITICAL entities are rejected outright. The toxicity check is one of the seven-rule precedence chain evaluated for every trade.

5.2 Market State Interaction

During STRESSED market state, toxicity enforcement is tightened: the threshold for HIGH tier drops from 51 to 40, and the threshold for CRITICAL drops from 76 to 60. This provides additional protection during volatile conditions when toxic flow is most damaging to LPs.

5.3 Compliance Integration

Toxicity scores are available to the compliance-api for regulatory reporting. Entities flagged as CRITICAL may trigger enhanced KYC reviews under the compliance zone's AML policies. The scoring data provides regulators with quantitative evidence of market manipulation patterns.

5.4 LP Protection Metrics

Liquidity providers can view aggregate toxicity statistics for the pools they participate in, including the distribution of flow by toxicity tier, the estimated adverse selection cost per period, and the effectiveness of enforcement actions in reducing toxic flow.

06Prior Art Differentiation

07Implementation Roadmap

08Patent Claim