Chart-to-data extraction with vision-language models (VLMs) is increasingly evaluated on benchmarks that show diminishing headroom (frontier VLMs exceed 89% on ChartQA) and with metrics that treat extracted points as unordered key-value pairs, ignoring the temporal structure of time series and penalizing small alignment shifts as catastrophic failures. We address both gaps with EpiCurveBench, a benchmark of 1,000 epidemic curve images curated from real-world public-health sources, and EpiCurveSimilarity (ECS), an evaluation metric that aligns predicted and ground-truth series via dynamic programming, tolerating local temporal shifts and gaps while penalizing them proportionally. Evaluating nine methods—three frontier closed VLMs, four open VLM, and two specialized chart-extraction systems—we find the strongest model reaches only 52.3% ECS, and that ECS spreads the four general-purpose VLMs over a 25-point range where key-value metrics (RMS, SCRM) compress them into a 5-point band. We further validate ECS against four downstream epidemiological summary statistics, finding that higher ECS predicts smaller errors in total counts, peak timing, and peak magnitude, and higher growth-rate fidelity; across all four, ECS correlates 1.5–3.6× more strongly than Dynamic Time Warping, which lacks a gap penalty and therefore cannot distinguish a truncated prediction from a temporally faithful one. EpiCurveBench targets a high-impact public-health application—unlocking decades of outbreak data trapped in published figures—but the benchmark and metric apply directly to any structured time-series chart-extraction setting.

Many research areas rely on data from theweb to gain insights and test their methods.However, collecting comprehensive researchdatasets often demands manually reviewingmany web pages to identify and record relevantdata points, which is labor-intensive and sus-ceptible to error. While the emergence of largelanguage models (LLM)-powered web agentshas begun to automate parts of this process,they often struggle to ensure the validity of thedata they collect. Indeed, these agents exhibitseveral recurring failure modes—including hal-lucinating or omitting values, misinterpretingpage semantics, and failing to detect invalidinformation—which are subtle and difficultto detect and correct manually. To addressthis, we introduce the AI Committee, a novelmodel-agnostic multi-agent system that auto-mates the process of validating and remediatingweb-sourced datasets. Each agent is special-ized in a distinct task in the data quality assur-ance pipeline, from source scrutiny and fact-checking to data remediation and integrity val-idation. The AI Committee leverages variousLLM capabilities—including in-context learn-ing for dataset adaptation, chain-of-thought rea-soning for complex semantic validation, and aself-correction loop for data remediation—allwithout task-specific training. We demonstratethe effectiveness of our system by applyingit to three real-world datasets, showing that itgeneralizes across LLMs and significantly out-performs baseline approaches, achieving datacompleteness up to 73.3% and precision up to97.3%. We additionally conduct an ablationstudy demonstrating the contribution of eachagent to the Committee’s performance. Thiswork is released as an open-source tool for theresearch community

Researchers across many fields rely on web data to gain new insights and validate methods. However, assembling accurate and comprehensive datasets typically requires manual review of numerous web pages to identify and record only those data points relevant to specific research objectives. The vast and scattered nature of online information makes this process time-consuming and prone to human error. To address these challenges, we present a human-in-the-loop framework that automates web-scale data collection end-to-end using large language models (LLMs). Given a textual description of a target dataset, our framework (1) automatically formulates search engine queries, (2) navigates the web to identify relevant web pages, (3) extracts the data points of interest, and (4) performs quality control to produce a structured, research-ready dataset. Importantly, users remain in the loop throughout the process and can inspect and adjust the framework’s decisions to ensure alignment with their needs. We introduce techniques to mitigate both search engine bias and LLM hallucinations during data extraction. Experiments across three diverse data collection tasks show that our framework greatly outperforms existing methods, while a user evaluation demonstrates its practical utility. We release our code at https://github.com/tberkane/web-data-collection to help other researchers create custom datasets more efficiently.