These runs are very likely also going to be a part of Reinforcement Learning and internal distillation no? As well as being analysed for what reasoning directions worked well, and which directions did not work well.

The mathematical results are valuable as PR, but the real value might be in the improvement this can bring to other training runs.