I don’t imagine a model capable of the first part would require being told not to assume a decipherment is correct