Stage-Audit: Auditable Source-Frontier Discovery for Cross-Wiki Tables

2026-05-19

AI Summary

This work addresses the challenge of large language models generating row-level data in cross-Wiki table construction without verifiable source support. To mitigate this issue, the authors propose the first auditable framework that enforces separation of write permissions between curators and auditors, introduces a row-level source citation gating mechanism, and defines a comprehensive set of 12 audit categories spanning coverage keys, schema alignment, and source roles. This design ensures that every generated table row is explicitly grounded in and traceable to credible sources. Experimental evaluation on a benchmark of 51 instances demonstrates substantial improvements: source-front precision increases by 42% (from 0.356 to 0.505) and F1 score rises by 35% (from 0.334 to 0.451), significantly enhancing both source accuracy and overall auditability.

Abstract

LLM-curated tables can appear source-grounded while containing unsupported rows: the curator may recall entries from parametric memory and retroactively attach page-level citations that are not the actual source. We study this hazard in Seed2Frontier discovery: the task of finding complement Wikipedia pages from a seed page to assemble a structured table. Stage-Audit addresses it with disjoint curator-auditor write rights, a row-level source-citation gate, and a 12-check audit taxonomy over keys, schema, source roles, cardinality, and scope. On a curated 51-instance Seed2Frontier evaluation set spanning 15 top-level domains, Stage-Audit improves source-frontier precision over a vanilla LLM curator from 0.356 to 0.505 (+42% relative) and F1 from 0.334 to 0.451 (+35%), while maintaining explicit per-row source traceability. The vanilla-LLM-vs-Stage-Audit comparison isolates the policy contribution rather than LLM-based discovery in general.

Problem

• source-groundedness
• cross-wiki tables
• Seed2Frontier discovery
• citation accuracy
• auditable curation

Innovation

• Stage-Audit
• source-grounded curation
• row-level citation
• auditable table construction
• Seed2Frontier discovery