BAP-SQL: Budget-Aware Observation Planning for Agentic Text-to-SQL

2026-08-03

AI Summary

This work addresses the inefficiencies in agent-based text-to-SQL tasks, where blind query execution often wastes contextual and database resources, and post-hoc compression struggles to recover critical information. To mitigate these issues, the study introduces budget-aware planning into SQL generation for the first time, dynamically rewriting queries based on risk assessment and integrating a standalone runtime safeguard. This approach enables policy-interpretable observation planning and budget-sensitive remediation under constrained resources. The method is compatible with language models of varying scales (4B/7B) and achieves a 3.4–3.6 percentage point improvement in accuracy over supervised fine-tuned baselines on a BIRD-derived benchmark, while simultaneously reducing token consumption by 4.5–5.0%.

Abstract

Tool-using agents do not merely consume observations: their actions determine what arrives next. In agentic text-to-SQL, a broad query can spend context and database work before useful evidence appears, while post-hoc compression cannot recover omitted rows or expended work. We present BAP-SQL, which treats observation formation as a budget-control stage: it estimates query risk, rewrites SQL when useful, and delegates hard limits to an independent runtime shield. Across general 4B, specialized FINER-SQL 4B, and 7B backbones, BAP-SQL improves tight-budget success. On the primary BIRD-derived setting, it gains 3.4/3.6 percentage points over matched SFT while using 4.5/5.0% fewer tokens. Matched retraining and task-level transfer associate the gain with policy-visible planning and budget-sensitive rescue. The benefit attenuates as model capability and budget increase, reverses at the loosest setting, and does not reduce database work.

Problem

• text-to-SQL
• observation planning
• budget-aware
• tool-using agents
• resource efficiency

Innovation

• Budget-aware planning
• Agentic Text-to-SQL
• Observation formation
• Query rewriting
• Runtime shielding