{"id":47069633,"title":"Show HN: I just shipped the canonical neuro-symbolic control demo","points":1,"user":"anulum","time":1771472584,"time_ago":"12 hours ago","type":"link","content":"

Quick update on the SCPN Fusion Core: \nI just shipped the canonical neuro-symbolic control demo.

One notebook. Zero setup. Runs in <2 min.

What it shows:\n- Stochastic Petri net → verified SNN compiler (LIF + bitstream path)\n- Formal topology/liveness/boundedness proofs (with SHA256 proof bundle)\n- Closed-loop on *real DIII-D shot 166000* (beta-limit disruption precursor)\n- Full FusionKernel digital twin evolution under SNN control\n- Side-by-side: SNN vs PID vs MPC — RMSE, disruption flags, actuator commands, latency (sub-ms p95)\n- Deterministic artifact export → replay with identical states/actions/proofs

Notebook (executed outputs baked in, Colab button too):\nhttps://lnkd.in/eb-vSv4r

Full repo + RESULTS.md (honest metrics, DIII-D/SPARC validation, limitations section):\nhttps://lnkd.in/eTJMfWC8

I’m pruning the kitchen-sink modules this week (legacy/ folder incoming) so it becomes a clean control-only package.

Would love brutal feedback:\n- Does the formal verification approach look credible for real-time safety?\n- Anyone at DIII-D National Fusion Facility-D / Princeton Plasma Physics Laboratory (PPPL) / ITER Organization willing to throw more shots at it?\n- What’s the biggest red flag you see for actual hardware-in-loop?

AGPL, fully reproducible, happy to hop on a call or add features.

Cheers

hashtag#FusionEnergy\nhashtag#NuclearFusion\nhashtag#AI\nhashtag#MachineLearning\nhashtag#Tokamak\nhashtag#PlasmaPhysics\nhashtag#ControlSystems\nhashtag#Neuromorphic\nhashtag#SNN\nhashtag#PetriNets\nhashtag#OpenSource\nhashtag#EnergyTech","url":"https://github.com/anulum/scpn-fusion-core/blob/main/examples/neuro_symbolic_control_demo.ipynb","domain":"github.com","comments":[],"comments_count":0}