the final RMSD is very high

I tried to use rfd3,mpnn and rf3 to make a demo.
First, i used rfd3 to redesign the sequence of A179-230 with cotig parameter.
The output use the ligandmpnn to redesign the sequence in A179-230 and resultec in "DTARFD3, id=1, T=0.1, seed=111, overall_confidence=0.5297, ligand_confidence=0.5339, seq_rec=0.3500
"
It seems very low.
Finally, i used the json file as below:
{
"name": "dimer_PLP_MN_complex",
"components": [
{
"seq": "RGIALPPAAQPGDPLARVDTPSLVLDLPAFEANLRAMQAWADRHEVALRPHAKAHKCPEIALRQLALGARGICCQKVSEALPFVAAGIRDIHISNEVVGPAKLALLGQLARAAKISVCVDNAENLAQLSAAMTRAGAEIDVLVEVDVGQGRCGVSDDATVLALAQQARALPGLRCVGVDDSARAVPGLRTVGGGTGSVEFDAASGVYTELQAGSYAFMDSDYGANEWNGPLKFQNSLFVLSTVMSTPAPGRVILDAGLKSTTAECGPPAVYGEPGLTYAAINDEHGVVRVEPGAQAPALGAVLRLVPSHVDPTFNLHDGLVVVKDGVVQDVWEIAARGFSR",
"chain_id": "A"
},
{
"seq": "GIALPPAAQPGDPLARVDTPSLVLDLPAFEANLRAMQAWADRHEVALRPHAKAHKCPEIALRQLALGARGICCQKVSEALPFVAAGIRDIHISNEVVGPAKLALLGQLARAAKISVCVDNAENLAQLSAAMTRAGAEIDVLVEVDVGQGRCGVSDDATVLALAQQARALPGLNFAGLQAYHGSVQHYRTREERAAVCRQAARIAASYAQLLRESGIACDTITGGGTGSVEFDAASGVYTELQAGSYAFMDSDYGANEWNGPLKFQNSLFVLSTVMSTPAPGRVILDAGLKSTTAECGPPAVYGEPGLTYAAINDEHGVVRVEPGAQAPALGAVLRLVPSHVDPTFNLHDGLVVVKDGVVQDVWEIAARGFSR",
"chain_id": "B"
},
{
"ccd_code": "PLP",
"chain_id": "C"
},
{
"ccd_code": "PLP",
"chain_id": "D"
},
{
"ccd_code": "PLP",
"chain_id": "E"
},
{
"ccd_code": "PLP",
"chain_id": "F"
},
{
"ccd_code": "MN",
"chain_id": "G"
},
{
"ccd_code": "MN",
"chain_id": "H"
}
]
}

and the predicted structure is not very good, with the score as below:
{
"chain_ptm": [0.69,0.7,0.6,0.59,0.59,0.59,0.61,0.6],
"chain_pair_pae_min": [
[null,19.69,19.33,17.34,11.66,12.59,12.78,12.45],
[null,null,9.1,13.19,17.95,17.28,12.25,12.64],
[null,null,null,16.05,18.24,18.66,12.66,12.56],
[null,null,null,null,18.34,18.23,12.99,12.78],
[null,null,null,null,null,17.27,13.14,12.83],
[null,null,null,null,null,null,12.9,13.01],
[null,null,null,null,null,null,null,12.37],
[null,null,null,null,null,null,null,null]
],
"chain_pair_pde_min": [
[null,8.49,8.51,7.87,6.38,6.62,10.98,11.1],
[null,null,4.71,6.55,8.14,7.91,10.9,10.63],
[null,null,null,5.13,6.75,6.59,12.44,11.08],
[null,null,null,null,6.47,6.81,12.86,11.92],
[null,null,null,null,null,6.26,12.31,12.46],
[null,null,null,null,null,null,12.1,13.35],
[null,null,null,null,null,null,null,15.57],
[null,null,null,null,null,null,null,null]
],
"chain_pair_pae": [
[null,25.8,24.87,24.08,21.4,22.09,17.17,16.93],
[null,null,19.46,21.65,24.33,24.17,16.66,16.95],
[null,null,null,19.78,23.19,23.25,16.05,16.0],
[null,null,null,null,22.68,23.24,16.3,16.23],
[null,null,null,null,null,21.27,16.37,16.12],
[null,null,null,null,null,null,16.17,16.25],
[null,null,null,null,null,null,null,12.54],
[null,null,null,null,null,null,null,null]
],
"chain_pair_pde": [
[null,11.62,10.45,10.2,9.13,9.6,12.67,12.77],
[null,null,7.79,9.17,10.24,10.25,12.57,12.34],
[null,null,null,7.16,8.25,8.35,13.45,12.05],
[null,null,null,null,8.16,8.46,13.75,12.65],
[null,null,null,null,null,8.35,13.2,13.41],
[null,null,null,null,null,null,12.75,14.16],
[null,null,null,null,null,null,null,15.57],
[null,null,null,null,null,null,null,null]
],
"overall_plddt": 0.6909,
"overall_pde": 9.3398,
"overall_pae": 21.6883,
"ptm": 0.4506063461303711,
"iptm": 0.3327164351940155,
"has_clash": false,
"ranking_score": 0.3563
}

The RMSD between model_0.cif (predicted by rf3) RFD3 output is 20.434.
Is there any problem in my workflow? The demo used structure (PDB 4v15)

Start by reproducing the described RFD3 → MPNN → RF3 workflow with the 4v15-derived JSON input and inspect model_0.cif. Compare the reported RMSD and confidence metrics at each stage; done means determining whether the workflow or input setup explains the discrepancy, with the findings documented.