Design a privacy-preserving federated learning pipeline for heterogeneous edge devices that includes client selection, secure aggregation, differential privacy, model versioning, and resource-aware training scheduling. Specify system architecture, algorithms/protocols (with rationale), privacy budget and DP mechanism, secure aggregation protocol and threat model, client selection and fairness strategies for non‑IID and unreliable clients, model versioning/compatibility and rollback procedures, resource-aware scheduler that accounts for compute, memory, battery, and network variability, and evaluation metrics and tests. Responses will be evaluated on correctness, feasibility, handling of heterogeneity, and privacy–utility trade-offs.