RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

arXiv 2607.25886 isolates data-centric research capability by fixing the entire post-training stack so only the agent's data strategy varies. Four frontier agents across six benchmarks. Among searches that continued past the best observed score, 78.26% ended on a lower-scoring final attempt and the rest merely recovered the peak. The practical read: your self-improvement loop needs explicit checkpoint preservation, not more iterations. Code at github.com/evolvent-ai/RSIBench-Data.