Search papers, labs, and topics across Lattice.
The authors argue that agent fragility under distribution shift and unfaithful explanations are symptoms of a shared structural flaw: training on observational interaction logs that inherently lack the counterfactual variation required to learn causal invariances. To resolve this, they formulate Data-Centric Anchoring and the Data-Centric Agentic Loop鈥攁 four-stage pipeline of Curation, Augmentation, Constraint, and Attribution designed to engineer causal structure directly into the training environment. Grounded in a taxonomy linking failure modes like miscalibration and spurious reliance to data lifecycle gaps, the work provides a blueprint for converting post-deployment failures into targeted counterfactual data interventions.
Scaling and post-hoc interpretability hit a hard wall in agentic AI because models cannot recover causal invariances that observational interaction data never contained in the first place.
Agentic AI systems built on large language models fail in two persistent ways that scaling does not fix: they break under distribution shift, and they cannot explain the decisions they make. We argue these are co-symptoms of one structural deficiency in the data lifecycle that governs how agents are trained, evaluated, and deployed. Observational interaction logs record what an agent did, not what it would have done otherwise. They encode spurious correlations without controlled variation, so they lack the counterfactual structure needed to separate causal signal from coincidence or to validate an explanation. No model-centric method can recover invariances the data never contained. We present Data-Centric Anchoring: robustness and interpretability should be engineered into the data environment, not extracted from models after training. Our central contribution is the Data-Centric Agentic Loop, a four-stage framework of Curate, Augment, Constrain, and Attribute. The ordering is structural, not stylistic. Curation precedes augmentation because generative models amplify whatever bias they are trained on. Augmentation precedes constraint because invariance objectives are vacuous without variation across environments to be invariant to. Attribution closes the loop, converting observed failures into targeted data interventions for the next iteration. Each stage manufactures the preconditions of the next, which makes the loop self-correcting rather than merely sequential. We ground the framework in a failure-driven taxonomy that links four core failure modes to the data lifecycle: spurious feature reliance, distribution-shift fragility, uncertainty miscalibration, and explanation unfaithfulness. We close with the limits of this approach and the open problems that stand between it and practical deployment at scale.
