Verifiable, Control-Oriented Learning On The Fly A Toy Problem