Instead of starting from zero and adding to it we could start from something small and working which learners adjust. This is possibly more realistic and it would also avoid problems implementing #285.
