In this paper, we study controlled adaptability of metamodel transformations. We consider one of the most rigid metamodel transformation formalisms -- automated grammar transformation with operator suites, where a transformation script is built in such a way that it is essentially meant to be applicable only to one designated input grammar fragment. We propose a different model of processing unidirectional programmable grammar transformation commands, that makes them more adaptable. In the proposed method, the making of a decision of letting the transformation command fail (and thus halt the subsequent transformation steps) is taken away from the transformation engine and can be delegated to the transformation script (by specifying variability limits explicitly), to the grammar engineer (by making the transformation process interactive), or to another separate component that systematically implements the desired level of adaptability. The paper lists and explains two kinds of different adaptability of transformation (through tolerance and through adjustment) and contains examples of possible usage of this negotiated grammar transformation process.