Temporal Leakage in Search-Engine Date-Filtered Web Retrieval: A Retrospective Forecasting Case Study

ACL ARR 2026 January Submission9482 Authors

Keywords: Retrospective Forecasting, Data Leakage, Information Retrieval, Evaluation Methodologies, Large Language Models, Retrieval-Augmented Generation (RAG), Web Search, Temporal Generalization, Benchmarking, Backtesting

Abstract: Search-engine date filters are widely used to enforce pre-cutoff retrieval in retrospective evaluations of search-augmented forecasters. We show this approach is unreliable: auditing Google Search with a before: filter, 71\% of questions return at least one page containing strong post-cutoff leakage, and for 41\%, at least one page directly reveals the answer. Using a large language model (LLM), gpt-oss-120b, to forecast with these leaky documents, we demonstrate an inflated prediction accuracy (Brier score 0.10 vs. 0.24 with leak-free documents). We characterize common leakage mechanisms, including updated articles, related-content modules, unreliable metadata/timestamps, and absence-based signals, and argue that date-restricted search is insufficient for temporal evaluation. We recommend stronger retrieval safeguards or evaluation on frozen, time-stamped web snapshots to ensure credible retrospective forecasting.

Paper Type: Short

Research Area: Resources and Evaluation

Research Area Keywords: evaluation methodologies, benchmarking, retrieval-augmented generation, reproducibility, fact checking, rumor/misinformation detection

Contribution Types: NLP engineering experiment, Data resources, Data analysis, Position papers

Languages Studied: English

Submission Number: 9482