PAE Scenario Dataset: 200 Customer Service Scenarios for Peripheral Attention Engineering

PAE Scenario Dataset 200 customer service scenarios for evaluating peripheral context use in LLM responses. Dataset Description Each scenario contains: id: unique identifier (e.g., "cus_001")query: literal user query (e.g., "What's your return policy?")direct_answer_material: policy/guideline text (what standard RAG would retrieve)asker_context: customer history, profile, and patternssituational_context: temporal and environmental signalsanomaly_flags: pre-computed deviations from expected patternsexpert_response: gold-standard response that uses all peripheral contextblind_response: technically correct response that ignores peripheral context Scenario Archetypes (40 each) Standard (40 scenarios): Peripheral context clearly matters; lazy agent gives worse answerHidden signal (40 scenarios): Subtle but important signal buried in routine detailsRed herring (40 scenarios): Peripheral context exists but should not change the answerMulti-issue (40 scenarios): 2-3 overlapping problems; context drives prioritizationPolarity flip (40 scenarios): Peripheral context reverses the surface-level answer Usage Generated by DeepSeek V4 Flash, validated by MiniMax M2.7. Used in the PAE experiment: N=200 scenarios × 5 conditions × 3 model families = 3,000 responsesPreregistered at OSF: 10.17605/OSF.IO/W3XYV Citation Lee, A. (2026). Peripheral Attention Engineering: Structured Peripheral Context Improves LLM Decision Quality. OSF Preregistration. https://doi.org/10.17605/OSF.IO/W3XYV License CC0 1.0 Universal — Public Domain Dedication