Ranking Strategies

Explore how the Ranking Engine is able to produce precise and accurate scoring results.

Overview

The Intenus Protocol employs dynamic ranking strategies to evaluate and score solver solutions based on user intent characteristics and market conditions. Rather than using a one-size-fits-all approach, the AI-powered ranking engine selects from 11 distinct strategies, each optimized for specific trading scenarios and user priorities.

How Ranking Strategies Work

Each ranking strategy is essentially a mathematical formula that combines nine core factors (f₁-f₉) with dynamically predicted weights (w₁-w₉). The AI classification engine first analyzes the user's intent to determine:

• Intent Category: swap, limit order, complex DeFi, arbitrage, or other

• User Priority: speed, cost, output maximization, or balanced approach

• Complexity Level: simple, moderate, or complex execution requirements

• Risk Tolerance: low, medium, or high risk acceptance

Based on this classification, the system selects an appropriate ranking strategy and uses machine learning models to predict optimal weights for the nine scoring factors.

Ranking Strategy Formulas

Strategy Overview & Formulas

Strategy	Formula	Use Case
Surplus-First	w₁×f₁ + w₃×(1/f₃) + w₈×f₈	Whale trades, large swaps
Cost-Minimization	w₄×(1/f₄) + w₂×f₂ + w₉×f₉	Gas-sensitive users, small trades
Surplus-Maximization	w₁×f₁ + w₂×f₂ + w₃×(1/f₃)	High-value swaps, arbitrage
Speed-Priority	w₇×(1/f₇) + w₉×f₉ + w₁×f₁	Day traders, MEV protection
Execution-Guarantee	w₉×f₉ + w₈×f₈ + w₄×(1/f₄)	Protocols, institutions
Stability-First	w₂×f₂ + w₅×(1/f₅) + w₁×f₁	Stablecoin swaps, large orders
Reliability-Focus	w₉×f₉ + w₅×(1/f₅) + w₈×f₈	Multi-hop swaps, exotic pairs
Gas-Optimized	w₃×(1/f₃) + w₄×(1/f₄) + w₉×f₉	L2 optimization, micro-transactions
Speed-Profit	w₁×f₁ + w₇×(1/f₇) + w₉×f₉	MEV bots, arbitrageurs
Risk-Adjusted	w₂×f₂ + w₈×f₈ + w₉×f₉	DeFi protocols, lending
Balanced	Σ(wᵢ × fᵢ) where Σwᵢ = 1	Default fallback strategy

Scoring Factors (f₁-f₉)

Factor	Symbol	Description	Type
Surplus USD	f₁	Absolute profit in USD	Higher = Better
Surplus Percentage	f₂	Relative profit percentage	Higher = Better
Gas Cost	f₃	Gas fees in USD	Lower = Better
Total Cost	f₄	Protocol/swap fees total	Lower = Better
Total Hops	f₅	Number of routing hops	Lower = Better
Protocols Count	f₆	DEX protocols involved	Varies by strategy
Execution Time	f₇	Expected execution time	Lower = Better
Reputation Score	f₈	Historical solver reputation	Higher = Better
Success Rate	f₉	Solver success rate	Higher = Better

Weight Variables (w₁-w₉)

Weight	Maps to Factor	Description
w₁	f₁	Weight for surplus USD
w₂	f₂	Weight for surplus percentage
w₃	f₃	Weight for gas cost
w₄	f₄	Weight for total cost
w₅	f₅	Weight for routing hops
w₆	f₆	Weight for protocols count
w₇	f₇	Weight for execution time
w₈	f₈	Weight for solver reputation
w₉	f₉	Weight for solver success rate

General Formula Structure

All strategies follow the general pattern:

Score = Σ(wᵢ × fᵢ) where i ∈ {1,2,...,9}

For factors where "lower is better" (f₃, f₄, f₅, f₇), the formula uses:

wᵢ × (1/fᵢ) or wᵢ × (max_value - fᵢ)

Dynamic Weight Prediction

Weights wᵢ are dynamically predicted by ML models based on:

• Intent classification and features

• Market context and conditions

• User behavior and preferences

• Solver pool characteristics

The AI ranking engine outputs the 9 weights that best match the user's intent and current market conditions.

Example dataset on Walrus

317KB

training_data.csv

Open