Root Cause Identification Engine

The engine builds a Directed Acyclic Graph (DAG) representing the supply chain's data-generating mechanism. Nodes represent entities like suppliers, ports, and inventory levels, while directed edges encode known causal relationships. This graph serves as the foundational map for all subsequent reasoning, distinguishing it from simple anomaly detection dashboards.

• Ingests Structural Causal Models (SCMs) defined by domain experts
• Integrates with Causal Discovery Algorithms to learn graph structures from observational data
• Continuously updates edges based on new operational data

When a disruption is detected—such as a sudden drop in on-time deliveries—the engine applies Do-Calculus rules to query the causal graph. It mathematically determines which variables must be controlled for to isolate the true origin, blocking backdoor paths and avoiding collider bias that would mislead a naive correlation analysis.

• Transforms interventional queries into estimable observational expressions
• Identifies sufficient adjustment sets to deconfound the analysis
• Handles latent confounders by flagging unidentifiable queries

The engine performs counterfactual reasoning to answer 'what would have happened if...' questions. For a late shipment, it estimates whether the delay would have occurred if a specific supplier had not experienced a raw material shortage. This isolates the root cause node by comparing the factual outcome against a simulated counterfactual world.

• Computes Individual Treatment Effects for specific disruption events
• Uses Structural Equation Modeling to simulate alternative scenarios
• Distinguishes direct causes from downstream ripple effects

A single disruption often has multiple contributing factors. The engine decomposes the total observed effect into direct, indirect, and interaction effects using mediation analysis. It quantifies how much of a delivery delay is attributable to port congestion versus the carrier's specific operational failure, preventing misattribution.

• Applies Mediation Analysis to trace causal pathways
• Identifies Heterogeneous Treatment Effects across different lanes or product categories
• Ranks contributing factors by their estimated causal contribution

The engine operates on streaming data, continuously evaluating causal invariance to distinguish genuine structural breaks from expected volatility. When a previously stable causal relationship shifts—indicating a fundamental change in the supply chain's behavior—the engine triggers an alert and re-evaluates the causal graph.

• Monitors for violations of Causal Invariance across environments
• Integrates with Supply Chain Control Towers for live visualization
• Uses Causal Impact Analysis on time-series telemetry to detect intervention points

The engine generates human-readable audit trails that justify why a specific node was identified as the root cause. It outputs the exact causal path, the variables conditioned on, and the estimated effect size. This algorithmic explainability is critical for risk managers who must validate the logic before taking corrective action.

• Visualizes the activated causal path within the DAG
• Reports the Average Treatment Effect and confidence intervals
• Provides natural language summaries of the counterfactual reasoning process