Phase A — provision_slice.ipynb

Provisions an aggregator plus N client nodes (mapped to ISO/RTO footprints) on a FABNetv4 routed network, installs PyTorch (GPU wheels by default, CPU with USE_GPU=false) and Flower, uploads the fl/ package, and writes results/slice_info.json. It does not run federated learning and does not poll — once the slice is READY, open federated_learning.ipynb.

Phase B — federated_learning.ipynb

Runs the federated learning against the ready slice. Launch star (Flower FedAvg/FedProx) or mesh (decentralized peer-averaging over FABNetv4) training, apply tc netem perturbation or mid-training churn, collect metrics.json, and render convergence / topology / communication figures. Re-run any cell with different settings on the same slice — no re-provisioning.

See weave.md for the full spec.