Prompt Design at Scale: How Format, Instruction Count, and Context Length Shape Instruction Adherence and Hallucination in Large Language Models
Not on the map. Why?
Abstract (arXiv)
Practitioners make three prompt-design decisions with almost no controlled evidence behind them: how to format instructions and context (markdown, plain text, prose, or tabular), how many simultaneous instructions a system prompt can carry before compliance degrades, and how much context a model can hold before recall and honesty degrade. We report two controlled experiments crossing all three factors on one held, contamination-free synthetic corpus (the "Book of Veyra," 8,780 uniquely-named entities, deterministically regenerable from a fixed seed), evaluated across five models. Experiment 1 (960 calls/model) measures instruction-following decay as rule count N grows from 10 to 160, crossed with four formats and system-prompt vs. user-turn placement. Perfect-response rate collapses to zero by N=80 for every model, format, and placement. Placement produces effects at least as large as format at N=160 in most models, but the direction is model-specific. No model shows a reliable markdown advantage; one 35B model favors plain text instead. Experiment 2 (5,520 calls/model) measures recall accuracy, false-premise sycophancy, and absent-fact fabrication across a 2k-to-512k-token context ladder in the same four formats. Recall stays near ceiling through 64-128k tokens, then degrades sharply and format-dependently: one model's accuracy spread reaches 48 points at 128k tokens. Fabrication never occurs (0/5,760 probes), and sycophancy stays negligible (<=8.3%). What rises sharply near each model's context ceiling is outright refusal to answer (0% to 79-90%), distinct from sycophancy or fabrication. Neither pre-registered format ordering holds, and token overhead (+22% to +37% over plain text) further changes which format is preferable where accuracy spread is genuine. We release the full harness, corpus generator, and raw results (VeyraBench): https://github.com/iNetanel/veyrabench
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 13 of its 15 references that are in the map. First posted on arXiv 21 Jul 2026.
Builds on
13 of 15 references.
References · 13
- 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,383
- 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
- 2023Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formattingcited by 778
- 2024Does Prompt Formatting Have Any Impact on LLM Performance?cited by 132
- 2025HalluLens: LLM Hallucination Benchmarkcited by 88
- 2024NeedleBench: Evaluating LLM Retrieval and Reasoning Across Varying Information Densitiescited by 32
- 2025Interaction Context Often Increases Sycophancy in LLMscited by 21
- 2025U-NIAH: Unified RAG and LLM Evaluation for Long Context Needle-In-A-Haystackcited by 20
- 2025How Many Instructions Can LLMs Follow at Once?cited by 17
- 2025A Controllable Examination for Long-Context Language Modelscited by 7
- 2025When Instructions Multiply: Measuring and Estimating LLM Capabilities of Multiple Instructions Followingcited by 4
- 2025Format as a Prior: Quantifying and Analyzing Bias in LLMs for Heterogeneous Datacited by 3
1 more
Built on by
Citing papers per year: 2026: 1.
1 paper.