Liechtenstein .li dataset seeks fine-tuning feedback

The pitch is interesting for niche, high-trust regional AI workloads, but the real value will depend on access terms, reproducibility, and external validation of the QA claims.

• –A 28M-token corpus is small for broad pretraining but can be practical for domain adaptation, instruction tuning, and retrieval-heavy assistants.
• –Strong provenance metadata (URL/timestamp/digest/offset) is a real differentiator for auditability and regulated deployments.
• –The multilingual mix with heavy German coverage makes it potentially useful for DACH legal/government assistants and cross-lingual retrieval.
• –Since this is currently a feedback request (not a fully published benchmarked release), teams will likely want a sample plus eval results before committing.