Large language models have shown strong performance across a range of legal tasks, but existing benchmarks rarely evaluate the ability to take and defend a legal position, reason under incomplete information, or synthesize multiple statutory provisions. This gap is particularly pronounced for Canadian law, which remains underrepresented in legal NLP. We introduce GRACE (Grounded Reasoning Adversarial Canadian IEgal examples), a dataset of 1,915 question–reasoning–answer instances grounded in Canadian federal legislation. GRACE covers three reasoning modes: adversarial advocacy, uncertainty, and applied reasoning. We develop a pipeline that partitions raw statutory text, generates scenario-based questions and reasoning, and filters examples through model-free citation verification and LLM-based quality auditing. As a proof of concept, we fine-tune CLeAR-4B (Canadian Legal Adversarial Reasoning), a lightweight model for grounded legal reasoning, and evaluate it against the unmodified Qwen3-4B base model in open- and closed-book settings. CLeAR-4B substantially improves agreement with teacher outputs and statutory citation behavior when the relevant act text is provided, while its grounding degrades sharply when the statute is withheld. These results suggest that GRACE can support the development of lightweight legal models that reason more effectively from supplied statutory text.

Dialogue systems typically optimize the immediate quality of a response, but rarely anticipate how that response will be interpreted by the user. This is especially important in emotional support conversations, where a locally plausible reply may still be perceived as judgmental, dismissive, pressuring, or misaligned with the user’s current need. We frame this as a problem of candidate response evaluation under user-side interpretation uncertainty. We propose Cognitive Look-ahead Simulation (CLS), an inference-time framework that previews likely user interpretations before committing to the final response. CLS expands single-pass generation into a Generate–Simulate–Evaluate–Revise pipeline: it first infers a Structured User-State Representation, then uses a Cognitive World Model as an operational one-step predictor of the user’s immediate inner thought and next-turn state under a candidate reply, and finally applies selective revision to high-risk drafts. We evaluate CLS on the SAGE benchmark using simulator-based outcomes and LLM-as-a-Judge capability profiling, with complementary human evaluations. Across multiple backbones, CLS improves dialogue outcomes and reduces severe failures compared with direct generation and self-refinement baselines. These results suggest that anticipating user interpretations provides a useful inference-time signal for identifying and revising high-risk responses in emotional support conversations.