Tag
TRACE-Bench proposes a compositional operator framework and benchmark to diagnose multi-reference image generation capabilities by decomposing tasks into atomic operations for detailed evaluation.
This paper proposes CPSAINT, a seven-layer integrity decomposition for agentic AI systems, paired with FRIESA-K, a residual-risk functional based on absorbing Markov models. It bridges structural failure path analysis with quantified risk estimates, demonstrating the framework on warehouse robotics and financial-services agent scenarios.