This work constructs OADD-Bench by treating empirical papers as records of schema in use, and evaluates lexical and neural retrieval, adapted schema-linking systems, and large language model (LLM) OADD agents.
Abstract
Researchers often approach a data repository with an abstract concept and ask which columns can measure it. Useful columns may not resemble the query; they may matter only as complementary indicators in a defensible measure. This need differs from schema linking and column retrieval, which begin from more explicit needs and reward direct relevance. We define operationalization-aware data discovery (OADD): given a broad question and a database, optionally under a scope constraint, OADD jointly determines how focal concepts can be measured with available data and identifies supporting columns. Developing OADD methods requires examples for design and evaluation, but asking researchers to supply conceptual questions and their columns is impractical. We construct OADD-Bench by treating empirical papers as records of schema in use. A question miner extracts and reframes a paper-supported question; a paper-conditioned column miner reconstructs its measurements and grounds them to database identifiers. We admit only mappings supported by the publication and database documentation. OADD-Bench contains 160 questions from 111 papers and 4,682 question-column labels. Each target records a measurement used in published research; the paper supplies the precedent, while the miners extract and ground it. We evaluate lexical and neural retrieval, adapted schema-linking systems, and large language model (LLM) OADD agents. Each method receives only a question, permitted years, and dataset metadata; source papers are used only to construct and document benchmark labels. At the largest output limit, direct retrieval reaches at most 0.185 recall. The strongest schema-linking adaptation reaches 0.401 but remains optimized for a different objective; an OADD-directed agent performs best at 0.465. Even this agent covers less than half the target columns, showing that OADD remains an open problem.
This work recast five text-to-SQL datasets as retrieval tasks at both table and column granularity, covering realistic and enterprise-scale schemas under two document representations, and proposes corpus-adaptive fine-tuning, establishing schema linking as a standalone retrieval task and lightweight, label-free corpus adaptation as a practical route to deploying it at enterprise scale.
Qingcheng Zeng, Puxuan Yu, Aman Mehta et al.· 0 citations
How users actually use LLM-powered data systems is examined: how they struggle to specify and validate queries, how they iterate extensively to converge on bespoke operators specific to their documents and domains, and what these patterns imply for the systems and benchmarks their community should be building.
Shreya Shankar, Aditya G. Parameswaran, U. Berkeley· 0 citations
Doc2DB-Bench is introduced, a benchmark for Document-to-Database construction, containing 203 long-document instances across 42 schemas and seven domain groups, with 117 entity tables, 132 relationship tables, 7,341 rows, and 41,935 cells, which provides a testbed for reliable, auditable, and relationally faithful LLM-based data systems.
Zhuowen Liang, Zhengxuan Zhang, Jiayang Wang et al.· 0 citations
LLM-generated search queries are widely used to augment IR evaluation, yet they may contain concepts that presuppose answer-side document knowledge, violating the information-access boundary of pre-search users. Existing validation metrics, including overlap, diversity, and effectiveness, cannot distinguish rare human-tail variation from candidate answer-side intrusion. We introduce concept provenance, a framework that assigns query concepts to backstory-supported, human-central, human-tail, and candidate answer-side zones, operationalizing a boundary that retrieval metrics alone cannot detect. Applying concept provenance to 77,004 queries across 100 UQV100 topics, 8 LLMs, and 5 prompt conditions with two extraction pipelines, we obtain a cross-pipeline token-HCIR Spearman rho of 1.0 over five condition means. Candidate answer-side concepts constitute 7.40 percent of non-generic concepts and appear in 97 of 100 topics, with topic explaining approximately 67 percent of variance. Human validation yields 68.2 percent relaxed precision, revealing two mechanisms: knowledge intrusion at 45.5 percent and deployment intrusion at 45.0 percent. Diagnostic probes show disproportionate localized retrieval effects, with deletion effect size d = -0.47 compared with d = -0.34 for random deletion, but these concepts explain less than 2 percent of aggregate evaluation variance. Concept provenance therefore serves as a boundary-compliance diagnostic rather than an evaluation-shift predictor. Under the tested conditions, no prompt condition eliminates intrusion; post-generation concept-provenance selection achieves 99 percent elimination.
Chenglong Ma, Xinye Wanyan, Danula Hettiachchi et al.· 0 citations
A framework that turns a traceable, reproducible, scope controlled research corpus into ranked, falsifiable research questions is presented, instantiate the framework on exoplanet atmospheres, a domain that uniquely combines literature, structured catalogs, and space telescope archives.
Text-to-SQL benchmarks ship schemas whose column names already say what the columns mean. Production warehouses are the inverse: cryptic identifiers, partial or absent documentation. We address the problem they pose first: recovering what columns and values mean from the data itself. Rosetta places a language model inside a verification harness: a deterministic profiler extracts structural evidence (value fingerprints, a 26-pattern library, checksum verdicts), the model proposes semantics conditioned on that evidence, and every fact carries provenance and a confidence bounded by its evidence class. Against human documentation on 680 paired columns across eleven BIRD databases, identifiers destroyed, the harness delivers metadata that is 0.475 accurate on the 42% of columns it commits to, against 0.223 on 94% for the same model used directly. Restricted to the 283 columns where both arms speak, the harness writes no better prose than the model alone; the gain is selection: deterministic evidence governs whether the system speaks (coverage +0.257 [0.128, 0.378]), not how well. The deterministic layer is a competence detector, not a competence amplifier. A backbone swap bounds the claim: the prose finding reproduces, but prompt-requested abstention does not transfer; a code-enforced commit gate (predictions registered first; measured on a third backbone and held-out databases) makes no-evidence coverage 0.000 on every backbone. On a blind i2b2 clinical warehouse Rosetta decodes 95.5% of 134 real ICD-9 codes from values alone and abstains on all 44 NDC drug codes. The catalog supports calibrated abstention at query time: under full schema opacity a naive translator falls from 0.92 to 0.42 execution accuracy while our gate answers at 86% accuracy over 59% coverage. Negative results are reported plainly, including that our own authority ladder is not the mechanism behind the headline.
