New TrAC algorithm enumerates all actual causes in causal models
An arXiv paper introduces a sound and complete procedure for finding every actual cause of an outcome, a task the authors say previously lacked such a method.
In plain words
The paper tackles actual causation: formally deciding which events in a specific situation caused a given outcome, using structural causal models. The authors present TrAC, an algorithm they prove lists all such causes rather than just one. They also report tests on random Boolean models with up to 15 nodes.
What was announced
The authors, writing on arXiv, describe Halpernian actual causation as the formal study of which events in a specific situation caused a specific outcome in acyclic structural causal models, a framework they say underpins legal responsibility, moral blame, and the assessment of causal harm. They state that no prior sound and complete enumeration procedure was known for finding all actual causes in a given setting, which they argue blocked downstream tasks that aggregate over the full cause set, such as responsibility attribution, blame assessment, and causal-harm grading.
Their technical contribution combines witness targets, defined as sets of joint configurations containing every counterfactual witness, with an ancestor intervention grammar to derive a finite directed acyclic witness map. They prove that for every actual cause there is a contingency set whose joint intervention appears as a node of this map, and they present TrAC, an algorithm they prove is sound and complete for enumerating all actual causes of any target event. Empirically, they compare TrAC with existing single-cause baselines on random Boolean structural causal models and report practical feasibility for graphs of up to 15 nodes.
What it does not show
- The evidence is the paper's abstract only; the full proofs, algorithm details, and experimental setup were not available for review.
- The scalability claim rests on the authors' own experiments on random Boolean models of up to 15 nodes and has not been independently confirmed.
- The paper reports a comparison with single-cause baselines, so no results against other full-enumeration methods are described.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.