Self-Consistency in Chat: Raising Reasoning Reliability Without Retraining

How multi-sample reasoning and aggregation can stabilize outputs in math, logic, and policy-heavy prompts.