Checkpoints are definitely a thing that could be implemented, but you then need to either trust that the software developers have put a valid (or honest) check

Checkpoints are definitely a thing that could be implemented, but you then need to either trust that the software developers have put a valid (or honest) checkpoint into the software... otherwise you are back to trying to determine consensus among hostile actors. A simple >95% consensus on a checkpoint is not enough when it effectively costs fractions of a cent to create thousands or millions of nodes that could all claim to have the "most correct" checkpoint.