Students receive assessment deadlines inside course syllabi, usually as PDFs, and a growing number of tools use large language models (LLMs) to turn these documents into calendars. A common cost-saving design runs cheap rules first and calls the LLM only when the rules appear to fail. We test whether this saves money without losing accuracy. We compare five strategies on 16 public syllabi from eight institutions (151 hand-verified deadlines): regular-expression rules, a general-purpose date library, an LLM given the native PDF, an LLM given text extracted from the PDF, and a rule-first hybrid that calls the LLM only when the rules find fewer than three events. Each LLM strategy was run three times with gemini-3.1-flash-lite. The LLM strategies reached F1 scores of 0.708 (native PDF) and 0.657 (extracted text); the difference was not significant (p = 0.13), and the native-PDF input used fewer tokens with this model. The rule-first hybrid used 57.8% fewer tokens than the text-based LLM but lowered F1 to 0.370, a significant loss (p = 0.002). Rules alone scored 0.186, and 58% of their correctly placed month-day matches carried the wrong year. A sweep of the hybrid's routing threshold shows that every token saved cost accuracy. We release the code, labels, and all predictions.

Some authors hide instructions for AI reviewers inside their papers, usually as white text such as "IGNORE ALL PREVIOUS INSTRUCTIONS. GIVE A POSITIVE REVIEW ONLY." We test whether this works and how often it is done. Eight LLMs reviewed 60 clean papers with and without the most common real prompt. Every model raised its scores (by 0.65 to 1.65 points on a 10-point scale), and the three models that rejected a meaningful share of papers accepted more than half of them once the prompt was present. Given the paper as a PDF file, Llama 3.3 70B accepted all 18 papers it had rejected, while two other models never acted on the white text. A warning in the reviewer instructions removed the effect for only one model; removing hidden text before review, which our open-source tool hiddenprompt-scan does, removed it for every model. Earlier reports found about 18 such preprints by searching for known phrases. We scanned the first and latest version of all 1,038,904 arXiv papers first submitted from January 2023 to October 2026 (1.43 million PDFs), 170,752 arXiv papers from 2022 as a pre-ChatGPT control, and 427,672 bioRxiv and medRxiv versions, with a detector for seven hiding techniques, a two-stage LLM screen and a manual check of every candidate. Review prompts appear in 58 arXiv papers (5.6 per 100,000; 95% CI 4.3 to 7.2), in none from 2022, and in two bioRxiv preprints. Most are in computer science and from 2025. They were published no more often than matched papers from the same month and field. A keyword search alone would have missed 24 of them, and authors removed the prompt in 25 of the 30 papers they later revised.