Operationalizing GDPR Art. 22 automated-decision explanations at scale

Our team has been handling DSARs that specifically invoke Art. 22 (right not to be subject to automated decision-making). The practical challenge is producing meaningful explanations for ML-based decisions that were never designed with interpretability as a first-class requirement. Current approach: we've built a post-hoc explanation layer using SHAP values + templated narratives, but the supervisory authority (LfDI Bayern) signaled in a recent audit that feature-importance scores alone may not satisfy the "meaningful information about the logic involved" standard. Questions for practitioners: - How did you structure your Art. 22 response templates to satisfy auditors? - Did you implement real-time explanation generation or pre-compute them per model version? - How do you handle cross-border DSARs where EU Art. 22 intersects with non-EU data processing (e.g. US-based inference endpoints)? Jurisdiction: DE, EU