Baselines

netforge_rl.baselines.

Policy
RandomPolicy	uniform MultiDiscrete([32, 100])
HeuristicBluePolicy	isolate first compromised host, else analyse
HeuristicRedPolicy	exploit without recon — usually 0 compromises
KillChainRedPolicy	recon → exploit → pivot; ~2–3 hosts/ep, SLA drops
jax_ppo	on-device IPPO (baselines/jax_ppo.py)

ExploitRemoteService needs prior DiscoverNetworkServices. Use kill-chain as the Red reference, not heuristic-red.

python -m benchmarks.build_leaderboard --episodes 5 --max-steps 150
python -m benchmarks.train_curve --name blue_ransomware --iters 40

Committed IPPO on ransomware: mean reward 0.06 → 0.71 (40 iters, CPU). Load with jax_ppo.load_params. RLlib: benchmarks/rllib_rmappo.py. Run · Self-play / Elo