As other posts have pointed out, one ‘model’ (as recognized by the tournament) can receive predictions from whatever ‘model’ (i.e., user’s computation) a user chooses that week. That provides a way around the difficulty and delay, involved in changing how much is staked on what (tounament-recognized) model.
This handy work-around would be even handier, if the newly-sourced model could be re-named. No big problem, if it cannot.