steps.md

🚀 KernelX: Final submission Runbook

Follow these steps in order to launch the full eBPF + AI scheduling stack.

🛰️ Step 1: Kernel Sentinel (Sensor)

Loads the eBPF probes into the Linux scheduler to begin 24D telemetry extraction.

cd kernel
sudo make load

🛠️ Step 2: Rust Bridge (Data Hub)

Reads eBPF ring buffer, manages Shared Memory (SHM), and saves training data.

# --record flag enables saving transitions to trajectories.json
cargo run --manifest-path bridge/Cargo.toml --release -- --record

🧠 Step 3: Strategist Brain (OpenEnv Server)

Starts the FastAPI environment and the LLM-based Grader (OpenEnv Compliance).

export PYTHONPATH=$PYTHONPATH:.
python3 -m brain.server.app

🎭 Step 4: Autonomous Loop (The Actor)

The scheduling loop that queries the policy and sends priority nudges to the Bridge via ZMQ.

export PYTHONPATH=$PYTHONPATH:.
python3 -m brain.server.run_autonomous --verbose --interval 50

🎨 Step 5: Mission Control HUD (Visualization)

Bt-op inspired TUI for the judges to see live telemetry, AI reasoning, and rewards.

cargo run --manifest-path ui/Cargo.toml

---

📈 Training Pipeline

If you have collected enough data in trajectories.json and want to retrain:

export PYTHONPATH=$PYTHONPATH:.
python3 training/run_pipeline.py --raw-data trajectories.json --output-root training_results