The Effect Propagation Process (EPP) rests on a single axiomatic foundation: causality is treated as a spacetime-agnostic monadic functional dependency in which one moment of a system arises from a previous moment through a transforming function. The classical static definition of causality is recovered as a specialization of this more general form.
Computable primitives
From this axiom a small set of computable primitives is derived: the Causaloid (a polymorphic container for causal structure), the Causal Monad (its sequencing counterpart), an explicit Context (the operational environment in which causality unfolds), and a Causal State Machine that links inference to action. An optional Effect Ethos provides a verifiable deontic safety layer.
Subsumption of classical methods
DeepCausality subsumes traditional causal inference methods — Pearl's structural causal models, Granger causality, dynamic Bayesian networks, the Rubin Causal Model, and CATE inference — as parametric specializations. Beyond that, DeepCausality enables dynamic causality as required for advanced science and engineering, as demonstrated in worked examples spanning avionics, chronometric geodesy, materials science, and relativistic, quantum, and condensed matter physics.
Implementation and governance
DeepCausality is implemented as a layered stack of production-quality Rust crates, released under the MIT open-source license and designed to be independently usable layer by layer. The project is hosted at the Linux Foundation for Data & AI under open governance.
Resources
- deepcausality.com — project home
- DeepCausality on GitHub — source, issues, discussions
- EPP monograph — preprint of the foundational paper
- Worked examples — cross-domain applications