Large language models (LLMs) excel at many natural language tasks, yet their reasoning reliability under structured perturbations of rule-based systems remains brittle. We present a controlled evaluation framework consisting of four stress tests: (1) rule deletion (redundant vs. essential); (2) contradictory evidence injection; (3) logic-preserving rewrites; and (4) multi-law equivalence stacking. While representative model families (BERT, Qwen2, and TinyLlama) achieve Acc = 1.0000 on base tasks, our framework reveals a critical failure mode termed Logic Inertia - a total breakdown (Acc = 0.0000) under contradictions, where deductive momentum overrides factual reality. To resolve this, we propose Conflict-Aware Fusion, a framework grounded in the Cognitive Structure Hypothesis which posits that robust reasoning requires an explicit structural inductive bias. By imposing a dual-process architecture that separates premise verification from logical deduction, Conflict-Aware Fusion eliminates logic inertia, achieving 1.0000 accuracy on both base and contradictory stress tests, and significantly enhancing robustness to missing evidence. Our results demonstrate that, for reliable multi-step reasoning, structural verification discipline is as critical as training data scale, providing a blueprint for building robust, contradiction-aware AI systems https://github.com/14H034160212/lemo. See the OpenAI/Evals pull request https://github.com/openai/evals/pull/1622.

翻译：大语言模型（LLMs）在众多自然语言任务中表现出色，但其在基于规则系统的结构化扰动下的推理可靠性仍然脆弱。我们提出了一个包含四项压力测试的受控评估框架：（1）规则删除（冗余与必要）；（2）矛盾证据注入；（3）逻辑保持重写；（4）多定律等价堆叠。虽然代表性模型家族（BERT、Qwen2和TinyLlama）在基础任务上达到准确率 = 1.0000，但我们的框架揭示了一种关键失效模式，称为逻辑惯性——在矛盾条件下完全崩溃（准确率 = 0.0000），此时演绎动量压倒了事实现实。为解决此问题，我们提出了冲突感知融合框架，该框架基于认知结构假说，该假说认为稳健推理需要明确的结构归纳偏置。通过采用将前提验证与逻辑演绎分离的双过程架构，冲突感知融合消除了逻辑惯性，在基础和矛盾压力测试中均实现了1.0000的准确率，并显著增强了对缺失证据的鲁棒性。我们的结果表明，对于可靠的多步推理，结构化验证规范与训练数据规模同等关键，为构建稳健、具备矛盾感知能力的人工智能系统提供了蓝图 https://github.com/14H034160212/lemo。参见OpenAI/Evals拉取请求 https://github.com/openai/evals/pull/1622。