1 paper
Yuxuan Lu, Ziyi Wang, Yingzhou Lu +12
Training tool-calling agents requires large-scale trajectory data with verifiable labels, yet existing approaches either synthesize environments that diverge from real API behavior…