It would be great to be able to work in a “debugging” mode. It could be as simple as blocking the release levers until they are manually triggered (so you could pause, analyze what the algorithm has done so far, and then move forward). This is an important skill and it would be great to be able to encourage pausing and tracking behavior. It would also be a nice setting to prevent accidental starting.
You could always just load one ball at a time. Once it traverses the board, load the appropriate color ball and manually release it.