{"slug": "dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning", "title": "Dual-Axis Policy Entailment: A Benchmark for Policy-Change Reasoning", "summary": "A new arXiv paper (2611.00536v1) introduces Dual-Axis Policy Entailment (DAPE), a benchmark of 1,000 expert-annotated ordered pairs of Chinese government policy directives that tests whether two directives govern the same policy domain and whether the later one continues, broadens, narrows or reverses the earlier one. Fourteen open-weight models and four frontier API flagships failed the task zero-shot, with the best frontier flagship reaching Cohen's κ ≈ 0.29 against a human ceiling of 0.53; a LoRA fine-tune of Qwen3-14B with five-fold cross-validation reached κ = 0.524 versus a coder-mutual human ceiling of 0.527. The authors apply the method to China's AI policy from 1980 to 2024 to map the direction of policy change across eras.", "body_md": "arXiv:2611.00536v1 Announce Type: new \nAbstract: Whether a policy shifts domains and, by extension, continues, broadens, narrows, or reverses previous policy is of paramount importance to the study of public policy. We propose Dual-Axis Policy Entailment (DAPE), which evaluates an ordered pair of policy directives on both dimensions at once: a binary judgment determines whether the two directives govern the same policy domain, and a five-way judgment determines whether the later directive continues, broadens, narrows, or reverses the earlier policy, or governs a different domain altogether. Read together, the two judgments form one five-category label, our primary metric. The two-axis design captures two central dimensions of policy change without conflating errors on either dimension into a generic \"irrelevant\" label, and it makes annotation more rigorous: a coder cannot dismiss a difficult pair as domain-irrelevant without an explicit, checkable decision. We introduce an expert-annotated benchmark of 1,000 ordered directive pairs drawn from a large corpus of Chinese government policy, together with an annotation-calibration protocol (an adjudication packet with pre/post agreement measurement) that we contribute as a methodological artifact. Fourteen open-weight models and four frontier API flagships fail the task zero-shot: even the best frontier flagship's chance-corrected agreement with expert labels is far below the human ceiling (Cohen's $\\kappa \\approx 0.29$ vs. $0.53$). A fine-tuned open-weight model (LoRA SFT of Qwen3-14B, five-fold cross-validation) closes this gap, reaching the coder-mutual human ceiling ($\\kappa = 0.524$ vs. $0.527$). Finally, we apply our method to China's AI policy (1980-2024) and map the direction of policy change across eras.", "url": "https://wpnews.pro/news/dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning", "canonical_source": "https://arxiv.org/abs/2611.00536", "published_at": "2026-11-04 05:00:00+00:00", "updated_at": "2026-11-04 05:21:19.410767+00:00", "lang": "en", "topics": ["ai-research", "large-language-models", "ai-policy", "natural-language-processing"], "entities": ["Dual-Axis Policy Entailment", "Qwen3-14B", "arXiv", "China"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning", "markdown": "https://wpnews.pro/news/dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning.md", "text": "https://wpnews.pro/news/dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning.txt", "jsonld": "https://wpnews.pro/news/dual-axis-policy-entailment-a-benchmark-for-policy-change-reasoning.jsonld"}}