DecoSearch: Complexity-Aware Routing and Plan-Level Repair for Text-to-SQL

2026-06-16

AI Summary

This work addresses the poor performance of large language models (LLMs) on complex, multi-step, and data-dependent Text-to-SQL tasks by proposing a training-free inference framework. The approach employs a lightweight schema selector to prune the database schema and a complexity-aware routing mechanism based on an LLM judge: simple queries are directly translated into SQL, while complex ones are decomposed into atomic subproblems structured as a directed acyclic graph (DAG). These subproblems are then resolved through retrieval-augmented generation (RAG) and topologically optimized for plan-level refinement. Evaluated on the BIRD and Spider benchmarks, the framework achieves execution accuracies of 70.53% and 88.31%, respectively—substantially outperforming existing training-free methods—while reducing inference token consumption by an order of magnitude. Moreover, it functions as a plug-and-play module that enhances the performance of existing SQL generation models.

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities in translating natural language to SQL, yet existing methods still falter on complex queries requiring multi-step, data-aware reasoning. We introduce DecoSearch, a training-free framework that addresses this by routing each query to the appropriate level of reasoning effort. A lightweight Schema Selector first prunes the full database schema to the relevant tables and columns. An LLM Judger then decides whether the question requires decomposition: straightforward questions follow a direct generation path and complex ones are escalated to a Directed Acyclic Graph (DAG) of atomic sub-questions, each solved by a targeted SQL generation step. A RAG component grounds the decomposer with semantically similar training examples, and a Topology Refiner restructures the reasoning plan when execution failures signal a flawed decomposition rather than a fixable SQL error. DecoSearch achieves 70.53% execution accuracy on BIRD and 88.31% on Spider with a DeepSeek backbone, surpassing all training-free baselines while consuming an order of magnitude fewer tokens than competing methods. It also functions as a model-agnostic wrapper, consistently improving fine-tuned SQL generation backbones without any modification to the pipeline.

Problem

• Text-to-SQL
• complex queries
• multi-step reasoning
• data-aware reasoning
• SQL generation

Innovation

• Complexity-Aware Routing
• Plan-Level Repair
• Decomposition DAG
• RAG-Guided Decomposition
• Training-Free Framework