RippleCP: Measuring Counterfactual Checkpoint Advantage in Coding Agents

Agent checkpoint systems decide what state is recovery-relevant, how to snapshot it, and whether rollback is admissible. None decides which of the safe boundaries they expose are worth materializing.