The transition to renewable energy grids exposes power systems to wind lulls lasting several weeks (Dunkelflaute). To ensure reliability, decarbonization strategies rely on long-duration energy storage (LDES) using hydrogen. However, deterministic models do not quantify the probabilistic risk of consecutive system failures. This paper proposes a hybrid framework that links a 34,944-hour PyPSA-Eur simulation of the 2030 German grid with a causal Bayesian Network. By translating the continuous grid physics into discrete states, we model compound failure mechanisms. Probabilistic inference reveals a critical nfrastructural bottleneck: during a power outage, the probability that hydrogen caverns are critically depleted is 65.5%, while the probability that gas-to-power fuel cells will be overloaded is 99.8%. This suggests that while volumetric energy targets (TWh) are necessary, they are structurally insufficient to prevent collapse without expanding energy conversion capacity (GW).