Wallet Graph Module (30%)
Counterparty concentration, wash patterns, insider propagation, and sybil pressure.
Credibility Anchor
How our risk score works.
We combine behavioral, structural, and market signals into one score. Every release is versioned so changes are visible and auditable.
MESS-RPE v0.3
Bi-weekly
Known Rugs + Stable Protocols
Top Signal Attribution
Score Breakdown
Composite score from 0-100 across five modules.
Liquidity Stability Module (25%)
Depth resilience, unlock pressure, and volatility stress projections.
Contract Semantics Module (20%)
Privilege paths, pause/mint controls, proxy risks, and hidden behavioral flags.
Tokenomic Simulation Module (15%)
Monte Carlo vesting shock scenarios and treasury concentration risks.
Behavioral Anomaly Module (10%)
Unsupervised detection against historical exploit and rug trajectories.
Versioning and Benchmarks
Versioned Releases
Every model update ships with changelog, data window, and validation deltas.
Every model update ships with changelog, data window, and validation deltas.
Benchmark Packs
Performance measured across balanced cohorts: known exploits, false-alarm controls, live protocols.
Performance measured across balanced cohorts: known exploits, false-alarm controls, live protocols.
Calibration Audits
Quarterly drift checks for precision, recall, and confidence calibration by score tier.
Quarterly drift checks for precision, recall, and confidence calibration by score tier.
Public Metrics
Precision, false positive rate, and top signal features exposed in dashboard telemetry.
Precision, false positive rate, and top signal features exposed in dashboard telemetry.
AI Technical Stack Design
Core modules, data sources, and public metrics used in the risk intelligence system.
A. Wallet Graph Intelligence
Graph Neural Networks (GNN), wallet clustering, insider ring detection, and sybil pattern recognition.
B. Liquidity Stability Model
Time-series forecasting (LSTM / Transformer), liquidity unlock simulation, and volatility risk projection.
C. Tokenomic Risk Simulation Engine
Monte Carlo simulations, vesting shock modeling, and concentration risk analysis.
D. Smart Contract Semantic Analysis
LLM-based contract parsing, malicious pattern detection, and upgradeability risk classification.
E. Behavioral Anomaly Engine
Unsupervised anomaly detection and historical rug-pull similarity scoring.
Data Sources
On-chain Transaction Data
Transfer history, entity behavior, and activity patterns.
Transfer history, entity behavior, and activity patterns.
Token Supply Data
Mint, burn, unlock, and distribution changes over time.
Mint, burn, unlock, and distribution changes over time.
DEX Liquidity Pools
Depth, volatility, and outflow/inflow behavior.
Depth, volatility, and outflow/inflow behavior.
Historical Exploit Datasets
Known rugs, exploit traces, and event signatures.
Known rugs, exploit traces, and event signatures.
Governance Activity
Proposal behavior, voting patterns, and control changes.
Proposal behavior, voting patterns, and control changes.
Public Transparency Metrics
Precision Rate
Measured accuracy of positive risk classifications.
False Positive Rate
Frequency of risk flags that do not materialize.
Historical Detection Accuracy
Backtested performance across labeled historical events.
Model Version Tracking
Public model versioning with changelog and evaluation deltas.
This methodology is an evolving framework. Final production weights and thresholds will be published with each stable release.