I have been trying to do that in Python, but it hits me that it was going to be really painful to chose the strategy on the fly. Well, I see how it could be done, but still painful. That’s why I tried pen&paper to chose the strategy : for 2-turns, it’s a reasonable task.
Well, it turns out that if yo go for the 5-reroll, you have *at least* ~9.7% to get 15 (assuming you chose the best strategy, which is keeping what you would keep normally), instead of a mere ~9.3%.
“At least” because I only counted the “good” first rerolls (not the “very good” nor the “bad”) : the ones that have exactly 3 or exactly 4 dices on , or . The true chances are then above those 9.7% but it shouldn’t be much more.
I’ll finish the programming part later ^^.