Published November 4, 2025
| Version 2
Publication
Open
Evans' Law: A Predictive Threshold for Long-Context Accuracy Collapse in Large Language Models
Description
Language models exhibit consistent performance decay as input and output lengths increase. This paper presents Evans' Law, defining the relationship between context length and accuracy degradation. Initial experimental validation confirms the phenomenon exists and provides empirical data to refine the mathematical formulation. Evans' Law: The likelihood of errors rises super-linearly with prompt and output length until accuracy falls below 50%, following a power-law relationship determined by model capacity and task complexity
Files
Evans Zenodo Submission.pdf
Files
(93.9 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:6a6ae8df226663c5dcb8990fa5966e5f
|
93.9 kB | Preview Download |
Additional details
Dates
- Created
-
2025-11-03
References
- Liu, N. F., et al. (2023). *Lost in the Middle: How Language Models Use Long Contexts.* Stanford CS. https://arxiv.org/abs/2307.03172 2. Zhang, Y., et al. (2025). *Context Length Alone Hurts LLM Performance Despite Perfect Retrieval.* arXiv:2510.05381 3. Veseli, B., et al. (2025). *Positional Biases Shift as Inputs Approach Context Window Limits.* arXiv:2508.07479 4. Chroma Research (2025). *Context Rot: How Increasing Input Tokens Impacts LLM Performance.* https://research.trychroma.com/context-rot (non-peer-reviewed industry report) 5. Evans, J. (2025). *Evans' Law: A Predictive Threshold for Long-Context Accuracy Collapse in Large Language Models.* [This paper]