A researcher utilized Astra to modify Qwen3.5-0.8B weights via Weight-Space Programming, achieving significant accuracy gains on specific tasks. In a demonstrated case, this method optimized a PPO network controlling a car in a Kinetix environment, successfully overcoming a local-minimum policy where the agent failed to reverse for sufficient momentum to scale a ramp.