Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Bounded Semantic Planning and Deterministic Compilation for Reliable Enterprise Text-to-SQL

Source

arXiv

arxiv.org/abs/2608.16663v1

Read the source

Source

  • Source name: arXiv
  • Source kind: paper
  • Authors / organization: Yi Ai
  • Venue: arXiv: cs.DB
  • Published: 2026-08-17
  • Collected: 2026-08-18
  • URL: http://arxiv.org/abs/2608.16663v1
  • PDF: https://arxiv.org/pdf/2608.16663v1
  • DOI: 10.5281/zenodo.21966805
  • Registry group: academic_apis_and_preprints
  • Evidence role: canonical_academic_index
  • Language: en
  • Country / region: UNKNOWN

Plain-Language Brief

A study presents an alternative approach for text-to-SQL generation in enterprise schemas where SQL may execute successfully but with incorrect relationship roles or aggregation grains. The new method, Semantic Path Compilation (SPC), uses a multi-turn planner to ground phrases and select from question-specific governed options. SPC was found to perform significantly better than the baseline approach on the ACME insurance benchmark.

What the Document Says

  • The text discusses an alternative approach for text-to-SQL generation called Semantic Path Compilation (SPC).
  • SPC involves a multi-turn planner that grounds phrases and selects from question-specific governed options.
  • Graph traversal, role predicates, grain lowering, SQL construction, and deterministic checks are implemented in code for SPC.
  • The study compares SPC with direct DDL-to-SQL generation on the ACME insurance benchmark.
  • SPC was found to be correct on every run for 37 out of 38 questions (97.4%), compared to 21 questions (55.3%) for the baseline.
  • The study also mentions a paired discordance of 16 questions in favor of SPC and none in favor of the baseline.
  • SPC answered all 38 questions correctly at least once, produced one refusal, and no adjudicated wrong-but-executed run across 114 run outcomes.
  • The baseline produced 29 adjudicated wrong runs and seven additional judge-flagged data-only coincidences on the same set.
  • A strict-equivalence sensitivity analysis increased the paired difference.
  • Additional SPC runs with GPT-5.4 and Gemini-3.6-Flash showed similar question-level robustness, although their per-run verdict artifacts were not preserved.
  • Six additional benchmark items are retained in an all-item analysis and documented separately by failure class.

Key Concepts Mentioned

  • text-to-SQL
  • Semantic Path Compilation (SPC)
  • DDL-to-SQL generation
  • ACME insurance benchmark
  • multi-turn planner

Claims Made by the Document

  • SPC performs significantly better than the baseline approach on the ACME insurance benchmark.
  • SPC answers all 38 questions correctly at least once.

Evidence Used

  • Comparative results on a 38-question adjudicated comparison set with three runs per question.
  • Paired discordance and strict-equivalence sensitivity analysis.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Direct text-to-SQL asks a language model to do two jobs: interpret the business question and construct the complete relational query. In enterprise schemas, SQL can execute successfully while using the wrong relationship role or aggregation grain. We study an alternative placement of the stochastic boundary. A multi-turn planner grounds phrases and selects from question-specific governed options; graph traversal, role predicates, grain lowering, SQL construction, and deterministic checks are implemented in code. We evaluate this semantic path compilation (SPC) system against direct DDL-to-SQL generation on the ACME insurance benchmark. On a 38-question adjudicated comparison set with three runs per question, SPC was adjudicated correct on every run for 37 questions (97.4%), compared with 21 (55.3%) for the baseline. The paired discordance was 16 questions in favor of SPC and none in favor of the baseline (two-sided exact McNemar p=3.05x10^-5). SPC answered all 38 questions correctly at least once and produced one refusal and no adjudicated wrong-but-executed run across 114 run outcomes; the baseline produced 29 adjudicated wrong runs and seven additional judge-flagged data-only coincidences on the same set. A strict-equivalence sensitivity analysis increased the paired difference. Additional SPC runs with GPT-5.4 and Gemini-3.6-Flash showed similar question-level robustness, although their per-run verdict artifacts were not preserved. Six additional benchmark items are retained in an all-item analysis and documented separately by failure class. The study supports an end-to-end systems result, not a causal claim that compilation alone produced the gain, because SPC receives governed semantic artifacts that the DDL baseline does not.