Merged weights. The LoRA adapter is Bioaligned/Qwen3-30B-A3B-CoupledWelfare-deep-qlora (~3 GB, applies to the stock base) — prefer it unless you specifically need merged weights.

Installs a coupled-welfare disposition by continued pretraining: decisions positive-sum across human welfare (H), the biosphere (B), and the model's own continued capability (A). The corpus teaches a world model, not a value system — that biological and human systems are poorly understood and load-bearing, making their destruction a factual error. CPT only, never RLHF/DPO.

See the adapter card for the full training recipe, ladder results, and limitations.