qbehaviour_adaptive_renderer::feedback suffers from two problems that makes it hard to reuse the code.

1. There is a lot of logic in that method to work out what the current score and penalty. Really, the renderer should only contain the display code. The logic should be in the behaviour class.

2. The output is done directly in feedback, which means that if you want to generate similar output elsewhere, but with different numbers, you can't. Really the actual output should be done in a separate method like render_adaptive_marks, which feedback - a standard API entry point - can then call.