Model Confidence & Anomalies
Risk assessment in DeFi is inherently uncertain.
Market conditions change rapidly, data quality varies across protocols, and extreme events are difficult to model precisely. For this reason, RAX separates risk estimation from confidence in that estimation.
What Is Model Confidence
Model Confidence represents how reliable a given risk assessment or simulation output is under current conditions.
It reflects the system’s internal certainty, not the level of risk itself.
A low-risk score with low confidence should be treated differently from a low-risk score with high confidence.
What Model Confidence Reflects
Model Confidence is derived from multiple factors, including:
Data completeness and freshness
Consistency across internal models
Stability of recent market behavior
Coverage of relevant scenarios
Presence or absence of anomalous signals
High confidence indicates that the system has sufficient, coherent information to support its assessment. Lower confidence indicates increased uncertainty and should be interpreted conservatively.
Why Confidence Is Separated from Risk
Risk and confidence answer different questions:
Risk Score asks: how risky is this relative to other options?
Model Confidence asks: how certain is this assessment?
Separating the two prevents false precision and allows users to make more informed decisions, especially during periods of market stress or structural change.
What Are Anomalies
An anomaly is a detected deviation from expected behavior.
Anomalies may arise from:
Sudden liquidity changes
Unusual volatility patterns
Abnormal yield movements
Inconsistent price or oracle data
Structural changes in protocol behavior
An anomaly does not necessarily indicate failure or loss. It signals that conditions differ from historical or modeled expectations.
How RAX Uses Anomalies
Anomalies are used to:
Increase caution in risk assessments
Adjust model confidence levels
Trigger alerts and monitoring workflows
Inform simulations and stress scenarios
Highlight areas requiring closer inspection
Anomalies act as early warning signals rather than conclusions.
Interpreting Low Confidence and Anomalies
When model confidence is low or anomalies are present, users should:
Avoid increasing exposure
Apply stricter allocation constraints
Rely more heavily on simulations and stress testing
Monitor conditions more frequently
Treat outputs as provisional rather than definitive
RAX is designed to surface uncertainty rather than hide it.
Relationship to the Allocation Engine
The Allocation Engine incorporates model confidence and anomaly signals when generating recommendations.
Lower confidence or active anomalies may:
Reduce recommended position sizes
Shift strategies toward more defensive profiles
Prevent execution of certain allocation changes
This ensures that uncertainty is explicitly reflected in capital decisions.
Summary
Model Confidence and Anomalies provide essential context for interpreting risk.
They allow RAX users to distinguish between stable, well-understood conditions and environments where uncertainty is elevated, enabling more cautious and disciplined decision-making.
Last updated