Causal Process Discovery

A family of techniques that infer causal execution dependencies — as opposed to merely temporal/frequency-based precedence — from process event logs. Distinguishes cause–effect directionality from correlation (Spirtes, Glymour, Scheines; Pearl).

Prior work in the BPM community (Hompes et al. 2017; Alaee, Weidlich, Senderovich 2024; Narendra et al. 2019; Leemans & Tax 2022 (all referenced-not-ingested)) typically extracts causal graphs from process KPIs or decision points. Fournier, Limonad, Skarbovsky, David 2025 (The WHY in Business Processes) infer causal graphs directly from activity timestamps by adapting LiNGAM-style discovery (Shimizu 2022).

In sources/2025-fournier-agentic-ai-process-observability the same machinery is applied to agent execution trajectories, producing an agent-centric causal view whose gateways reveal functional dependencies among tool invocations and split points marking behavioral variability.

Philosophical framing

"Causal" in causal process discovery carries all the philosophical commitments of concepts/causation. Timestamp-based discovery (LiNGAM, Granger-style tests) is closer to a regularity/probability-raising theory; functional dependency discovery on agent trajectories is closer to interventionist / mechanistic reasoning (concepts/mechanisms-in-science). No single method captures every symptom of causation, so plural methods are warranted (sources/2023-anjum-rocca-phi403-lecture-20-plural-methods-one-causation).

concepts/process-discovery · concepts/agent-process-observability · concepts/behavioral-variability · concepts/causation · concepts/mechanisms-in-science