Abstract
LLM analytics agents are evaluated on SQL syntax accuracy, but production failures look different: questions with two valid business definitions, questions the warehouse cannot answer, deprecated columns after a schema change, and queries that execute successfully while returning the wrong business number. No execution-match metric can score them. This paper introduces WarehouseReliabilityBench, 400 frozen tasks over two synthetic warehouses in which roughly half the correct responses are a clarification, an abstention or a refusal, with pinned denominators and a pre-registered paired bootstrap fixing each claim verb before the numbers existed. QueryProof, a 7B agent, uses rules derived from a semantic layer and physical catalog to determine its behaviour, and gates every answer on deterministic post-execution checks. On an 80-task synthetic test split evaluated once, QueryProof outperforms a direct-prompted 32B baseline by +0.237 [+0.112, +0.375] Business Truth Rate at 71.0% lower cost per correct answer; against a cost-matched few-shot baseline the accuracy gain holds but the cost difference does not resolve. This compares systems rather than model sizes: the 32B baseline receives none of the scaffolding. False success falls from 0.754 to 0.351 of returned answers, and no wrong number was returned on an answerable task (0 of 24), though 13 answers went to questions requiring clarification or abstention. Removing the routing layer changes little (0.562 against 0.537), so the result does not depend on escalation. Routing tuned on validation over-abstains on test, and the fitted confidence model loses to the heuristic it replaced. Resampling template families rather than tasks widens both accuracy intervals to include zero, so the effect's direction is better supported than its magnitude. The gain tracks the deterministic layer, though no component ablation was run.
Explore similar work
Apr 28, 2026cs.AI
LLMs deployed for natural-language querying of analytical databases suffer from two intertwined failures - incorrect answers and confident hallucinations - both rooted in the same cause: the model is forced to infer business semantics that the schema does not encode. We test whether supplying those semantics as context closes the gap. We benchmark three frontier LLMs (Claude Opus 4.7, Claude Sonnet 4.6, GPT-5.4) on 100 natural-language questions over the Cleaned Contoso Retail Dataset in ClickHouse, using a paired single-shot protocol. Each model is evaluated twice: once given only the warehouse schema, and once given the schema plus a 4 KB hand-authored markdown document describing the dataset's measures, conventions, and disambiguation rules. Adding the document improves accuracy by +17 to +23 percentage points across all three models. With it, the three models are statistically indistinguishable (67.7-68.7%); without it, they are also indistinguishable (45.5-50.5%). Every cross-cluster comparison is significant at p < 0.01. The presence of the semantic-layer document accounts for essentially all of the significant variance; model choice within tier does not. We interpret this as a structural result: explicit business semantics suppress the dominant class of text-to-SQL errors not by making the model more capable, but by changing what the model is being asked to do.
Michael Rumiantsau, Ivan Fokeev
Jul 7, 2026cs.LG
Evaluating uncertainty in AI-generated SQL queries requires estimating whether a query is correct, where correct means it executes to the same result as a human-written reference. We study which signals predict correctness on hard multi-table text-to-SQL, using AUROC to measure how well each ranks correct queries above incorrect ones. On BIRD and Spider, black-box signals such as string, structural, and execution self-consistency, a schema-relevance score, and query executability all fall between about 0.61 and 0.68 AUROC, with string self-consistency strongest at 0.675; white-box log-probability is similar (0.67). The signals that move past this ceiling are verification-based: an LLM judge scores from 0.72 (GPT-4o-mini) to 0.78 (Claude). Judges from different providers make different errors, so a two-provider ensemble reaches 0.82 AUROC with a well-calibrated probability (expected calibration error 0.03) and supports useful abstention frontiers (for example, answering 27% of questions at 24% selective risk) where self-consistency offers no valid low-risk subset. The pattern holds across two benchmarks, two generators, and two judge providers. We also ask whether a verifier can be trained. Fine-tuned verifiers, both encoder and generative, reach about 0.77 to 0.79 AUROC in-distribution but fall to about 0.66 on unseen schemas; scaling to 7B, adding schema diversity, distilling a strong judge's rationales, and cross-benchmark training all fail to close that gap. Cross-schema transfer appears to track model scale and reasoning rather than fine-tuning. In practice, correctness uncertainty for text-to-SQL lives in reasoning-based signals: a fine-tuned verifier is a good in-domain tool, but a verifier that generalizes across schemas currently means a large frozen reasoning model.
Robert Richardson
Apr 24, 2026cs.AI
LLM-based agents for text-to-SQL often struggle with latency-performance trade-off, where performance improvements come at the cost of latency or vice versa. We reformulate text-to-SQL generation within the lens of software test coverage where the original query is prepared with a suite of test cases with simpler, atomic SQLs that are executed in parallel and together ensure semantic coverage of the original query. After iterating on test case coverage, the final SQL is generated only when enough information is gathered, leveraging the explored test case SQLs to ground the final generation. We validated our framework on a state-of-the-art benchmark for text-to-SQL, Spider 2.0, achieving a new state-of-the-art with 70.2% execution accuracy.
Tanmay Parekh, Ella Hofmann-Coyle, Shuyi Wang +3