Vision-Language Models (VLMs) have demonstrated remarkable performance in document understanding tasks; however, VLMs also suffer from hallucinations inherited from LLMs. While prior work has focused on reasoning-stage hallucinations, the role of visual perception remains underexplored. In this work, we define perceptual hallucination as the phenomenon where VLMs generate information as if perceived, despite absent or damaged visual evidence. To analyze this, we construct DocHallu, a benchmark of 2,671 original–damaged image pairs across three tasks, available at https://huggingface.co/datasets/IB99/DocHallu. Experiments reveal that perceptual hallucination occurs across all models, with higher rates for numerical content than textual content. Activation patching analysis suggests that hallucinations are strongly associated with errors introduced in the vision encoder, which can subsequently propagate and become amplified through the text decoding process. We also demonstrate that LLM-based post-hoc filtering can reduce hallucination exposure by 36% on average, with reductions of up to 88%. This work extends VLM hallucination research by defining, analyzing, and verifying perceptual hallucination in document understanding.

Peer review is the primary mechanism for evaluating scientific contributions, yet prior studies have mostly examined paper features or external metadata in isolation. The emergence of open platforms such as OpenReview has transformed peer review into a transparent and interactive process, recording not only scores and comments but also rebuttals, reviewer–author exchanges, and meta-reviewer decisions. In this paper, we present a large-scale empirical study of ICLR 2017–2025, encompassing over 28,000 submissions, integrating four dimensions: paper structure and language quality, submission strategies and external metadata, author–reviewer interaction dynamics, and reviewer disagreement patterns. Our results show that factors beyond scientific novelty significantly shape acceptance outcomes. In particular, the rebuttal stage emerges as a decisive phase: timely, substantive author–reviewer communication strongly increases acceptance likelihood, often outweighing initial reviewer skepticism. Clearer writing, earlier submission, and effective disagreement resolution also correlate with higher acceptance probabilities. Based on these findings, we propose data-driven guidelines for authors, reviewers, and meta-reviewers to enhance transparency and fairness in peer review.