Partial diffusion returning the input molecule without modifications

I am trying to run RFD3 from the docker image on runpods. My input is a pdb of a receptor with a peptide ligand. I want to do partial diffusion, with some parts of the peptide fixed and others diffused in, while keeping the receptor conformation fixed. My json file is as follows:

{
"receptor_binder": {
"dialect": 2,
"infer_ori_strategy": "hotspots",
"input": "/workspace/receptor_clean.pdb",
"contig": "B1-368,3,/0,A1,2,A4-6,1,A8-9,1,A11,10,A22,2,A25-26,6",
"select_hotspots": {
"B256": "OE1",
"B102": "CD1",
"B254": "NH1"
},
"partial_T": 2
}
}
Inference appears to run, but the output models are all identical to the starting pdb file.

Start by reproducing the supplied RFD3 Docker run with the JSON configuration and input PDB. Compare generated models with the starting structure and trace partial-diffusion handling for the contig and partial_T settings. Done means the receptor remains fixed while the selected peptide regions are diffused rather than every output matching the input.