Take a look at the Position/Velocity vs Time view. What you should see is sort of a modulated sine wave, starting out small, getting big, getting smaller again, repeatedly with the amplitude of the size variations getting smaller as time goes on. What is happening is that the analytical solution p=A*sin(t) is only an approximation which is increasingly accurate as the amplitude A gets smaller. For large swings the period of the pendulum does depend on amplitude. Since the period of the pendulum varies while the period of the kicks does not, we get kicks occurring at various points in the cycle. Some of them even opposing the the motion of the pendulum. That is why the amplitude may decrease at times.