Distributed systems such as cloud-native applications and microservice architectures often experience cascading failures that are difficult to diagnose from observational data alone, including logs, traces, and metrics. Traditional Architecture Description Languages (ADLs) and runtime root-cause analysis (RCA) tools focus mainly on correlations and therefore do not distinguish spurious associations from true causal relationships, nor do they support counterfactual reasoning. C-ADL, presented in this paper, is an Architecture Description Language that integrates Pearl’s Structural Causal Models (SCMs) with component–connector software architectures. Building on prior work in causal observability, it embeds causal graphs directly into architectural descriptions at design time and makes intervention and counterfactual reasoning explicit in the architectural artifact itself. With C-ADL, architects can encode causal dependencies, evaluate do-interventions, and formulate counterfactual queries during design rather than only after deployment. We provide a formal metamodel, a review-friendly YAML-based concrete syntax, and automated tooling for validation, translation, and analysis. We evaluate the approach on three distributed systems derived from realistic benchmarks, including DeathStarBench, and show 83% Precision@1 for root-cause localization and 92% counterfactual prediction accuracy on verifiable cases.
