Ranking Engine
What is Ranking Engine?
The Core Orchestrator of Intenus
The Ranking Engine is Intenus's AI-powered system that evaluates competing solver solutions and selects the optimal one for execution. Unlike simple price comparisons, it performs sophisticated multi-dimensional analysis within a secure Trusted Execution Environment (TEE).
Why AI-Powered Ranking?
Traditional Systems use simple, deterministic metrics:
Highest output amount
Lowest gas cost
First-come-first-served
Intenus Ranking Engine considers complex, nuanced factors:
Historical solver reliability
Risk-adjusted returns
Market condition analysis
User preference patterns
Execution probability
How It Works
1. Intent Classification
When solutions arrive, AI models first classify the intent:
Input: "Swap 1,000 SUI for USDC, minimize slippage"
Classification:
- Type: swap.exact_input
- Complexity: simple
- Risk tolerance: low
- Priority: execution quality over speedThis classification determines which ranking strategy to apply.
2. Solution Scoring
Each solution is evaluated across multiple dimensions:
Economic Factors (40% weight):
Output amount vs. benchmark
Gas efficiency
MEV capture for user
Reliability Factors (30% weight):
Solver reputation score
Historical accuracy
Success rate for similar intents
Execution Factors (20% weight):
Expected completion time
Slippage probability
Failure risk
Strategic Factors (10% weight):
Novel approaches
Market adaptation
Innovation bonus
Ranking Strategies
Read more about Ranking Strategies Ranking Strategies /li
Different intent types use specialized strategies:
For Spot Swaps
Focus: Output maximization + reliability
Score = (0.4 × output_amount) +
(0.3 × solver_reputation) +
(0.2 × execution_speed) +
(0.1 × innovation)For Limit Orders
Focus: Execution certainty + price accuracy
Score = (0.5 × price_accuracy) +
(0.3 × execution_probability) +
(0.2 × gas_efficiency)For Complex Strategies
Focus: Successful completion + cost efficiency
Score = (0.4 × total_value) +
(0.3 × completion_risk) +
(0.2 × solver_reputation) +
(0.1 × gas_optimization)Multi-Factor Analysis
Economic Efficiency
Output Surplus:
surplus = (actual_output - minimum_expected) / minimum_expectedHigher surplus = better deal for users
Gas Optimization:
efficiency = expected_value / total_gas_costBalances value against execution cost
Solver Reputation
Reputation score aggregates:
Accuracy: Promised vs. delivered outcomes
Reliability: Success rate over last 100 intents
Consistency: Performance variance
Longevity: Time active in network
Reputation Formula:
reputation = (accuracy × 0.4) +
(reliability × 0.3) +
(consistency × 0.2) +
(longevity × 0.1)Risk Assessment
AI models evaluate:
Market volatility impact
Liquidity depth risks
Smart contract dependencies
Execution complexity
High-risk solutions require higher expected value to win.
Speed Metrics
Expected Execution Time:
Current network congestion
Transaction complexity
Historical solver speed
Protocol response times
Confirmation Probability:
Gas price adequacy
Transaction ordering likelihood
Network conditions
Verifiable Computation
TEE Infrastructure
Ranking occurs in Nautilus Trusted Execution Environment:
Security Properties:
Isolation
Code runs in secure enclave
Attestation
Cryptographic proof of correct execution
Confidentiality
Solution details remain private
Integrity
Results cannot be tampered with
Continuous Learning
Training Data Collection
Every execution generates training data:
Intent specification
Competing solutions
Chosen winner
Actual outcomes
Market conditions
Data is stored on Walrus for:
Model retraining
Strategy improvement
Performance auditing
Reinforcement Learning
AI models learn from outcomes:
Positive Reinforcement:
Winning solutions that performed as promised
Unexpected solver excellence
Novel successful strategies
Negative Reinforcement:
Solutions that underperformed
Failed executions
Inaccurate estimates
Model Updates
Updated models are:
Tested on historical data
Validated in testnet environment
Gradually rolled out
Monitored for performance
Fairness Guarantees
Preventing Bias
No Favoritism:
All solvers evaluated by same criteria
Reputation earned through performance
Stakes level playing field
Transparent Weights:
Ranking factors publicly documented
Weights adjusted through governance
Changes announced in advance
Verifiable Results:
TEE attestations prove correctness
Anyone can audit rankings
Disputes resolved with cryptographic evidence
Anti-Gaming Measures
Sybil Resistance:
Stake requirements limit fake identities
Reputation doesn't transfer between addresses
Economic cost to create multiple solvers
Collusion Detection:
Statistical analysis of submission patterns
Correlated behavior flagged
Coordinated solutions penalized
Quality Standards:
Minimum performance thresholds
Automatic disqualification for violations
Regular performance reviews
Performance Metrics
User Outcomes
The engine's effectiveness is measured by:
Average Surplus: How much better than direct execution
Execution Success Rate: Percentage of completed intents
User Satisfaction: Repeat usage patterns
Cost Savings: Gas + price improvements
Solver Distribution
Healthy competition indicators:
No single solver wins >30% of intents
New solvers can compete effectively
Innovation is rewarded
Multiple viable strategies exist
Technical Architecture
Components
Intent Classifier:
ML model trained on historical intents
Determines optimal ranking strategy
Runs inside TEE
Score Calculator:
Implements ranking formulas
Aggregates multi-dimensional data
Produces final scores
Proof Generator:
Creates cryptographic attestation
Packages ranking results
Enables verification
Data Archiver:
Stores results on Walrus
Feeds continuous learning
Enables auditing
Future Enhancements
Planned improvements:
Personalized Rankings:
Learn individual user preferences
Adapt to trading patterns
Optimize for specific goals
Predictive Analysis:
Forecast market movements
Optimize execution timing
Anticipate liquidity changes
Cross-Chain Intelligence:
Coordinate multi-chain strategies
Optimize bridge selection
Minimize total cost
Learn More
Last updated