BABAPPAΩ frozen model: reference weights for branch–site episodic selection inference

This archive contains the frozen TorchScript reference model (babappaomega.pt)used by BABAPPAΩ, a mechanistically grounded framework for likelihood-freebranch–site inference of episodic evolutionary deviation. The model was trainedon one million simulated codon alignments generated under a mutation–selectionframework incorporating recombination, epistasis, heterogeneous selectiveregimes, and alignment noise, and all parameters are fixed. The artifact isdistributed as an immutable TorchScript model and serves as the single referencemodel for BABAPPAΩ v1.2.x inference. Model weights are distributed separatelyfrom the BABAPPAΩ software package and are automatically downloaded, verifiedvia SHA-256 checksum, cached locally, and reused to ensure lightweightinstallation, transparent provenance, and bitwise reproducibility. This modelis intended for inference, benchmarking, and reproducible comparison under theassumptions described in the accompanying manuscript, and complements ratherthan replaces classical likelihood-based branch–site methods.