Good for All, Not Good Enough for One: Reuse Dilemma in Federated Learning

Federated learning (FL) promises privacy-aware collaboration in healthcare, but real-world adoption remains limited by infrastructural and organizational hurdles. In this paper, we reflect on our experience developing and later bypassing our own general-purpose FL framework, in favor of a task-specific pipeline. This case exposed five core barriers to reuse, ranging from workflow misalignment to governance constraints, that often go unaddressed in technical design. Rather than prescribing one approach over another, we argue for a shift toward modular, interoperable tools that can better accommodate the diversity of research contexts. Our findings highlight the need for realistic infrastructure thinking: one that acknowledges both the promise and the limits of reuse in practice.