The paper introduces a method for attributing changes in alarm scores in industrial and scientific systems by modeling the problem as a Shapley game over hybrid structural causal models. It distinguishes between changes in structural functions and exogenous laws, providing a detailed, function/law-resolved explanation of how each type of change propagates to the alarm score. For ordered linear SCMs and quadratic scores, the authors present a polynomial‑time exact computation of the full Shapley vector and, for scarce incident data, a cross‑fitted estimation approach with error bounds and top‑k recovery guarantees, validated through experiments on controlled and real shifts.
arXiv:2606. 08275v1 Announce Type: cross Abstract: When an LLM agent fails -- issues a refund it should not have, calls the wrong tool, leaks data -- existing tooling answers what happened (observability) or whether it passed (evaluation), but not which step caused the failure.
arXiv:2607. 27263v1 Announce Type: new Abstract: Most benchmarks for causal inference over time series are observational, small, or domain-specific, leaving interventional and counterfactual estimation under-served exactly where it matters most, such as in healthcare, policy evaluation, and climate science.
arXiv:2610.28803v1 Announce Type: new Abstract: While statistical learning has made remarkable progress in recent decades, driven by large datasets and data-driven protocols that have transformed sci...
arXiv:2608. 12555v1 Announce Type: new Abstract: Predictive explanation methods attribute a model output; they do not, by themselves, attribute an intervention effect on the real-world outcome.
arXiv:2607. 14940v1 Announce Type: new Abstract: We study causal inference under outcome interference for sequential, observational settings.
arXiv:2609.36881v1 Announce Type: new Abstract: Causal foundation models (CFMs) pre-trained on data generated from various structural causal models (SCMs) have been proposed for estimating causal eff...
The paper introduces Structural Causal Bottleneck Models (SCBMs), a new class of structural causal models where high‑dimensional causal effects are mediated by low‑dimensional bottleneck variables. SCBMs allow mechanism‑specific, target‑dependent dimensionality reduction while remaining estimable with standard learning algorithms. The authors prove that bottleneck variables are identifiable up to bijection from observational data, and demonstrate experimentally that using bottlenecks improves identification and finite‑sample efficiency in causal effect estimation for instrumental variables, mediation, and backdoor adjustment.
The paper presents a framework for causal attribution in agentic AI systems, outlining estimators and conditions where they fail. It distinguishes between marginal total effects and common‑random‑number total effects, introduces a natural direct effect under pinned downstreams, and derives a coupling method to keep direct effects estimable. The authors also propose a traceability specification to meet upcoming regulatory requirements for high‑risk AI systems.
arXiv:2606. 10703v1 Announce Type: new Abstract: Interpretability methods routinely use population-level summary statistics over observed model behaviour to license claims about the effects of targeted interventions on specific computations; in Pearl's terms, they treat rung-1 associational evidence as if it supported rung-2 interventional conclusions, a move whose validity is rarely tested.
arXiv:2609.36771v1 Announce Type: cross Abstract: Root cause analysis (RCA) is a critical problem in many real-world scenarios. RCA enables the identification of faulty or failing mechanisms in a sys...
arXiv:2610.00424v1 Announce Type: cross Abstract: Knowing how much a causal predictor could improve need not reveal the gain of the repair actually learned. We quantify this gap in a scalar Gaussian...
The paper introduces new one‑step and targeted minimum loss‑based estimators for the average treatment effect and the average treatment effect on the treated under the front‑door criterion. These estimators use multiple parameterizations of the observed data distribution, can avoid modeling the mediator density, and are compatible with flexible machine‑learning nuisance estimation. The authors derive conditions for root‑n consistency, asymptotic linearity, and develop doubly robust tests for identification assumptions, demonstrating improved efficiency and favorable performance in simulations and real‑world applications in education and emergency medicine.