We used Protein Hunter with Boltz-2, but replaced ProteinMPNN with fixed-backbone Caliby in the design pipeline. After obtaining a set of candidates with Boltz-2 ipSAE_asym (aligned on target, scored on binder) > 80, we predicted all complexes with AlphaFold3 (with MSA for target chain, no MSA for binder chain, and co-folded with 3 Zn2+). We then took all designs with AlphaFold3 ipSAE_asym > 80 and redesigned all sequences with ensemble-conditioned Caliby. Ensemble-conditioned Caliby refers to using partial diffusion with Protpardelle-1c to generate an ensemble of 32 backbones, followed by ensemble-conditioned sequence design conditioned on the target with SolubleCaliby. Both ensemble-conditioned designs and fixed-backbone designs were then predicted with binder-only and filtered for AlphaFold3 binder pLDDT > 80. We then bucketed by length and chose the best by AlphaFold3 ipSAE_asym. The final set of designs contains 50 fixed-backbone designs and 50 ensemble-conditioned designs.