partial diffusion is not suitable for protein binder design

When I used RFD3 to perform local optimization for protein binder design (within the context of generating a backbone), I observed that the structure of the target chain had changed, but the sequence remained unchanged.
How can I fix the whole target chian to miantain the structrue。
the blue chains are targets （the light one the original taget，the dark blue one is the one after diffusion）

The issue names RFD3 and local optimization during backbone generation but no file, entry point, configuration, or test. Start by reproducing the reported diffusion behavior and inspect how the target chain is handled during local optimization. Done means the target structure remains fixed while the intended binder design proceeds, with a regression test or documented reproduction result.