Hand-authored Natural Language Inference pairs for South Africa's Protection
of Personal Information Act (POPIA). Each example is a (premise, hypothesis,
label) triple covering one of 7 canonical POPIA clauses. Designed for
fine-tuning small cross-encoder NLI models into deterministic, local,
auditable compliance judges — the alternative to prompting an LLM-as-judge
over personal information.
Used to fine-tune labrat-aiko/nli-popia-v1,