Base language model for the Arwen Policy project — an evidence-grounded, multistakeholder AI deliberation system for digital policy and Internet governance.

This repository packages the Arwen Policy configuration for use with Qwen 3 via Ollama. The model is configured for evidence-grounded policy analysis with a 40K context window, low temperature (0.2), and a system prompt that instructs the model to attribute claims to documented sources, preserve stakeholder disagreements, disclose missing perspectives, and never manufacture consensus or evidence.

This is the base Qwen 3 model with an Arwen Policy system prompt and inference configuration. It has not been fine-tuned on the Arwen Policy Corpus.

For the task-adapted version, see Arwen Policy LoRA.

See CITATION.cff in the project repository.