More Than 'Plausible Nonsense': A Rigorous Eval for ADÉ, Our Security Coding Agent

Introduced the three-pillar evaluation framework — detection accuracy, robustness, and economic cost of coverage — that Sublime uses to move LLM-generated detection rules from “plausible-sounding” to provably correct, walking through how ADÉ’s generated rules are deconstructed and scored. This framework became the basis for the CAMLIS 2025 paper.

Read: Sublime blog post