Masking Stale Observations Helps Search Agents -- Until It Doesn't: A Regime Map and Its Mechanism

Benchmark Model Rank Results
deep-research-agents-on-browsecomp-plusDS-V4-Flash-Max + AgentIR-4B (stale-observation masking, K=5)#2Accuracy: 84.3
deep-research-agents-on-browsecomp-zhDS-V4-Flash-Max + Serper API (stale-observation masking, K=5)#2Accuracy: 73.4