SymbXRL Interactive Analysis

Probability Distribution of KPI Effects (Fig. 2)

Correlation Density Map — tx_brate

Decision Sequence Patterns (A → B → C) most frequent 3-step action chains

KPI State Transition Patterns P(next state | current state)

Policy & Explanation Consistency Over Time

Metric Definitions

Policy Consistency — weighted mean P(dominant next-state | current state). Measures how deterministic the agent's policy is: 100% means it always picks the same next decision from a given state; a drop signals the agent explores different actions from the same state.

Effect Consistency — weighted mean P(dominant KPI symbol | state→next-state edge). Measures how reliably a given decision transition produces the same KPI outcome: 100% means fully predictable effects; a drop signals environment non-stationarity or conflicting traffic regimes.

Explanation Stability — 1 − mean |Δedge probability| vs. the previous checkpoint. Measures how much the knowledge-graph edge probabilities shift over time: 100% means the graph has converged and explanations are reliable; a dip means the probabilistic view is still evolving.

Reading at t = —

Policy Consistency is —

Effect Consistency is —

Explanation Stability is —