ARXIV:2603.16448 · TEXT-TO-SQL · SUBMITTED 02 APR · 02:30 UTC · FRESHNESS STALE

VerifiedSource: PDF linkedPartialPaperPack: 3 of 4 citation fields filledMissingMissing fields: authorsPartialProof: unverified proof status

TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas

arXiv

TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

Blocked on Code›Score7.0Evidence unverified

Opportunity summary

Pain TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

Evidence 0 refs | 0 sources | 17% coverage

Blocker Evidence unverified

Open Build Read PDF Signal Canvas Track

PROBLEM

TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata. However, this premise fails in real-world enterprise environments where databases contain hundreds of tables with massive noisy metadata.

METHOD

Full abstract

Text-to-SQL parsing has achieved remarkable progress under the Full Schema Assumption. However, this premise fails in real-world enterprise environments where databases contain hundreds of tables with massive noisy metadata. Rather than injecting the full schema upfront, an agent must actively identify and verify only the relevant subset, giving rise to the Unknown Schema scenario we study in this work. To address this, we propose TRUST-SQL (Truthful Reasoning with Unknown Schema via Tools). We formulate the task as a Partially Observable Markov Decision Process where our autonomous agent employs a structured four-phase protocol to ground reasoning in verified metadata. Crucially, this protocol provides a structural boundary for our novel Dual-Track GRPO strategy. By applying token-level masked advantages, this strategy isolates exploration rewards from execution outcomes to resolve credit assignment, yielding a 9.9% relative improvement over standard GRPO. Extensive experiments across five benchmarks demonstrate that TRUST-SQL achieves an average absolute improvement of 30.6% and 16.6% for the 4B and 8B variants respectively over their base models. Remarkably, despite operating entirely without pre-loaded metadata, our framework consistently matches or surpasses strong baselines that rely on schema prefilling.

RESULT

ScienceToStartup currently rates this 7.0/10 on the public viability pass. Extensive experiments across five benchmarks demonstrate that TRUST-SQL achieves an average absolute improvement of 30.6% and 16.6% for the 4B and 8B variants respectively…

WHY NOW

Text-to-SQL moved forward this cycle; last verified April 2026. Public score 7.0/10.

Continue into Read for claims, analysis, references, and neighboring papers.

Opportunity summary

Score7.0

PainTRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

Evidence0 refs | 0 sources | 17% coverage

Blockermissing authors

Analysis summary

TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

VerifiedSource: PDF linkedPartialPaperPack: 3 of 4 citation fields filledMissingMissing fields: authorsPartialProof: unverified proof status

Competitive landscape

TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

Segment

Text-to-SQL

Adoption evidence

No public code link in the paper record yet

Commercial read

7.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

{ "contract_version": "paper-r2", "paper_id": "e12cddb8-de39-485d-8ffe-380d0c769fbc", "arxiv_id": "2603.16448", "canonical_route": "/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "active_tab": "synced from current hash by the drawer client", "selected_artifact": "trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "endpoints": { "paper_pack": "/api/v1/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas/paper-pack", "build_passport": "/api/v1/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas/build-passport", "mcp_resource": "sciencetostartup://surfaces/paper-workspace" } }

{ "surface": "paper", "mode": "paper", "query": "TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas", "normalized_query": "2603.16448", "route": "/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "paper_ref": "trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "topic_slug": null, "benchmark_ref": null, "dataset_ref": null }

{ "@context": "https://schema.org", "@graph": [ { "@type": "WebPage", "@id": "https://sciencetostartup.com/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas#webpage", "url": "https://sciencetostartup.com/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "name": "TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas", "description": "TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.", "isPartOf": { "@id": "https://sciencetostartup.com/#website" } }, { "@type": "ScholarlyArticle", "@id": "https://sciencetostartup.com/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas#scholarlyArticle", "headline": "TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas", "description": "TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.", "url": "https://sciencetostartup.com/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas", "sameAs": "https://arxiv.org/abs/2603.16448", "identifier": { "@type": "PropertyValue", "propertyID": "arXiv", "value": "2603.16448" }, "isAccessibleForFree": true, "isPartOf": { "@id": "https://sciencetostartup.com/#website" }, "datePublished": "2026-03-17T12:30:42.000Z", "additionalProperty": [ { "@type": "PropertyValue", "propertyID": "viabilityScore", "value": 7 }, { "@type": "PropertyValue", "propertyID": "researchDomain", "value": "Text-to-SQL" } ] }, { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://sciencetostartup.com" }, { "@type": "ListItem", "position": 2, "name": "Text-to-SQL", "item": "https://sciencetostartup.com/topics" }, { "@type": "ListItem", "position": 3, "name": "TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning", "item": "https://sciencetostartup.com/paper/trust-sql-tool-integrated-multi-turn-reinforcement-learning-for-text-to-sql-over-unknown-schemas" } ] }, { "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "What products could be built from this research?", "acceptedAnswer": { "@type": "Answer", "text": "The timing is right because enterprises are increasingly adopting self-service analytics tools but face limitations with existing text-to-SQL solutions that require clean, predefined schemas. The rise of large language models has created demand for more robust AI agents that can handle real-world data complexity. Market conditions favor solutions that reduce data team bottlenecks, as companies seek to leverage AI for operational efficiency amid economic pressures to cut costs and accelerate insights." } }, { "@type": "Question", "name": "What are the practical use cases?", "acceptedAnswer": { "@type": "Answer", "text": "A commercial use case is an AI-powered query assistant for a retail company's inventory database, which has over 500 tables with inconsistent naming conventions and missing metadata. Business analysts could ask natural language questions like 'What were the top-selling products in the Northeast region last quarter?' and the system would autonomously identify relevant tables (e.g., sales, products, regions), verify metadata, and generate the correct SQL query without pre-loaded schema knowledge, enabling faster inventory optimization and sales reporting." } } ] } ] }

Competitive landscape

TRUST-SQL revolutionizes Text-to-SQL parsing by enabling agents to dynamically identify relevant database schemas without pre-loaded metadata.

Segment

Text-to-SQL

Adoption evidence

No public code link in the paper record yet

Commercial read

7.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas

TRUST-SQL: Tool-Integrated Multi-Turn Reinforcement Learning for Text-to-SQL over Unknown Schemas

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Related Resources

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Related Resources

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline