QUIVER: A Formal Framework for Quantifying Perturbation Propagation and Bifurcation in Compound AI Systems
The article introduces QUIVER, a formal framework designed to quantify perturbation propagation and bifurcation in compound AI systems. This framework addresses the challenges posed by the stochastic nature of nodes and divergent execution paths in AI architectures. QUIVER has been validated on various production pipelines, demonstrating its ability to reveal sensitivity profiles and predict trajectory bifurcations.
- ▪QUIVER is a framework for measuring perturbation propagation in graph-structured LLM pipelines.
- ▪It defines a sensitivity matrix that classifies edges and identifies bifurcation thresholds.
- ▪The framework has been validated on two production enterprise pipelines and a public multihop QA pipeline.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23956 (cs) [Submitted on 11 May 2026] Title:QUIVER: A Formal Framework for Quantifying Perturbation Propagation and Bifurcation in Compound AI Systems Authors:Prashanti Nilayam, Sankalp Nayak View a PDF of the paper titled QUIVER: A Formal Framework for Quantifying Perturbation Propagation and Bifurcation in Compound AI Systems, by Prashanti Nilayam and 1 other authors View PDF HTML (experimental) Abstract:Compound AI systems that chain multiple LLM calls into directed computation graphs are now the dominant architecture for production AI.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
