# Instruction Stacking Collapse: A Benchmark and the Capability-Dependent Value of Prompt Compilation

arXiv: https://arxiv.org/abs/2608.02639 · HTML: https://mlsanity.com/paper/2608.02639/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 12 references that are in the map.
- First posted on arXiv 31 Jul 2026.

## Builds on (most cited first, 7 of 7)

- [Instruction-Following Evaluation for Large Language Models](https://mlsanity.com/paper/2311.07911/) (2023)
- [Inverse Scaling: When Bigger Isn't Better](https://mlsanity.com/paper/2306.09479/) (2023)
- [PRewrite: Prompt Rewriting with Reinforcement Learning](https://mlsanity.com/paper/2401.08189/) (2024)
- [LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints](https://mlsanity.com/paper/2410.06458/) (2024)
- [RECAST: Expanding the Boundaries of LLMs' Complex Instruction Following with Multi-Constraint Data](https://mlsanity.com/paper/2505.19030/) (2025)
- [Brevity Constraints Reverse Performance Hierarchies in Language Models](https://mlsanity.com/paper/2604.00025/) (2026)
- [On the Paradoxical Interference between Instruction-Following and Task Solving](https://mlsanity.com/paper/2601.22047/) (2026)

## Built on by (most cited first, 1 of 1)

- [How Should a Prompt Optimizer Spend a Tight Budget? BudgetAPO with Noise-Adaptive Evaluation](https://mlsanity.com/paper/2610.05671/) (2026)

## Abstract (arXiv)

Production prompts rarely carry a single instruction. One system message may require valid JSON, a word limit, three citations, and a fixed tone at the same time. We study how instruction-following degrades as such constraints accumulate. We introduce a benchmark that stacks 24 verifier-checked instructions, one to twenty at a time, and evaluate three production-tier LLMs (Claude Sonnet 4.6, GPT-5-mini, Gemini 2.5 Flash). Instruction-following degrades non-linearly: the follow rate falls from ~96% to as low as 20%, driven by a structured and reproducible set of pairwise conflicts. A single "output JSON" constraint, for example, is jointly unsatisfiable with nine others. We then evaluate a training-free remedy: an instruction compiler that rewrites the stacked prompt in a single LLM call and is reused across queries. Its benefit is capability-graded. It recovers up to +11 points of follow rate for weaker models, which are also the models most often deployed at scale, while leaving stronger models, which already internalise the same structure, essentially unchanged. Cluster-robust tests, same-baseline controls, and a within-family scaling ladder attribute the gain to the rewrite itself rather than to additional tokens, reordering, or measurement headroom. We release the benchmark, verifiers, and cached runs for full reproduction.