Schema linking is a pivotal step in Text-to-SQL pipelines, aiming to fetch the tables and columns relevant to a user query while discarding irrelevant ones. Imperfect linking often omits required columns, harming the correctness of the generated SQL.\ \ This work revisits schema linking in the context of the latest large language models (LLMs). Empirical results show that modern LLMs can implicitly identify and use the necessary schema elements during generation, even when the prompt contains many unrelated schema items. Consequently, when the entire schema fits within the model’s context window, our Text-to-SQL pipeline completely skips explicit schema linking, avoiding the risk of filtering out required elements.\ \ To further boost accuracy, we adopt three techniques:\
- Augmentation: Adding examples or extra descriptions in the prompt to help the model grasp the database structure;\
- Selection: Filtering candidate SQL statements before final output, keeping only those best aligned with the context;\
- Correction: Applying a post‑processing model or rule‑based fixes to rectify errors in the generated SQL.\ \ On the BIRD benchmark, this approach achieves a 71.83% accuracy, ranking first.\ \ Review: The paper demonstrates that, given a sufficiently large context window, powerful LLMs can replace explicit schema linking, simplifying the pipeline. Combined with augmentation, selection, and correction, this yields a substantial performance gain for Text-to-SQL systems.