Fatigue design for maritime structures currently relies on deterministic models of idealised geometries, which do not account for the substantial variability introduced by misalignments and imperfections. A probabilistic framework is presented that models imperfections as spatially varying random fields or random variables, using the Stochastic Hungry Horse model, Spectral Representation Method, Halton sampling, and Bayesian updating. Applied to a finite element model of a stiffened panel with Non-Watertight details, results show that local plate distortions reduce stresses under water pressure but increase them under global hull girder bending. Stiffener distortions govern the Hot Spots at the Non-Watertight details, producing both tensile and compressive stress concentration factors. Comparison with guidelines indicates that current safety factors are non-conservative, as secondary bending effects are not addressed. The framework captures fabrication- and assemblyinduced variability and supports data-driven refinement of fatigue criteria.