Academic grant programs are widely used to motivate international research collaboration and boost scientific impact across borders. Among these, bi-national funding schemes -- pairing researchers from just two designated countries - are common yet understudied compared with national and multinational funding. In this study, we explore whether bi-national programs genuinely foster new collaborations, high-quality research, and lasting partnerships. To this end, we conducted a bibliometric case study of the German-Israeli Foundation (GIF), covering 642 grants, 2,386 researchers, and 52,847 publications. Our results show that GIF funding catalyzes collaboration during, and even slightly before, the grant period, but rarely produces long-lasting partnerships that persist once the funding concludes. By tracing co-authorship before, during, and after the funding period, clustering collaboration trajectories with temporally-aware K-means, and predicting cluster membership with ML models (best: XGBoost, 74% accuracy), we find that 45% of teams with no prior joint work become active while funded, yet activity declines symmetrically post-award; roughly one-third sustain collaboration longer-term, and a small subset achieve high, lasting output. Moreover, there is no clear pattern in the scientometrics of the team's operating as a predictor for long-term collaboration before the grant. This refines prior assumptions that international funding generally forges enduring networks. The results suggest policy levers such as sequential funding, institutional anchoring (centers, shared infrastructure, mobility), and incentives favoring genuinely new pairings have the potential to convert short-term boosts into resilient scientific bridges and inform the design of bi-national science diplomacy instruments.

翻译：学术资助项目被广泛用于激励国际研究合作并提升跨国科学影响力。其中，双边资助计划——仅配对来自两个指定国家的研究人员——虽属常见，但与国家级及多边资助相比，其研究尚不充分。本研究旨在探讨双边项目是否真正促进了新的合作、高质量研究及持久伙伴关系。为此，我们对德以基金会（GIF）进行了文献计量案例研究，涵盖642项资助、2,386名研究人员及52,847篇出版物。研究结果表明，GIF资助在资助期间甚至略早于资助期催化了合作，但很少在资助结束后形成持久的伙伴关系。通过追踪资助前、中、后期的合著关系，使用时序感知K-means聚类合作轨迹，并利用机器学习模型（最佳模型：XGBoost，准确率74%）预测聚类归属，我们发现：45%先前无合作记录的团队在资助期间变得活跃，但资助后活跃度对称下降；约三分之一的团队维持了较长期的合作，而一小部分团队实现了高产且持久的成果。此外，团队在资助前的科学计量学特征并未显示出作为长期合作预测因子的明确规律。这修正了先前关于国际资助通常能建立持久网络的假设。研究结果提示，诸如连续性资助、机构锚定（中心、共享基础设施、人员流动）以及鼓励真正新配对的激励措施等政策杠杆，有潜力将短期促进转化为稳固的科学桥梁，并为双边科学外交工具的设计提供参考。