Diagnosis Is Not Prescription: Linguistic Co-Adaptation Explains Patching Hazards in LLM Pipelines
Quick Answer
The study reveals that correcting the routing module in LLM pipelines often worsens performance due to a Diagnostic Paradox, where upstream corrections yield better outcomes.
Quick Take
This is explained by the Linguistic Contract hypothesis, indicating that downstream modules adapt to upstream error distributions, and disrupting this alignment can degrade performance. Empirical evidence across three agent families supports this finding.
Key Points
- Routing module corrections often degrade performance in pipelines.
- Upstream query-rewriting module fixes consistently improve outcomes.
- The Linguistic Contract hypothesis explains the observed performance asymmetry.
- Higher co-adaptation correlates with patching harm across agent families.
- Findings are based on empirical analysis across three independent agent families.
Paper Resources
📖 Reader Mode
~2 min readAbstract:When a multi-module LLM agent fails, the module most responsible for the failure is not necessarily the best place to intervene. We demonstrate this Diagnostic Paradox empirically: causal analysis consistently identifies the routing module -- which selects which tool to call next -- as the primary bottleneck across three independent agent families. Yet injecting prompt-level correction examples into this module consistently degrades performance, sometimes severely. Patching an upstream query-rewriting module instead reliably improves outcomes. The effect holds with statistical significance on two agent families and directional consistency on a third; alternative repair strategies at the routing module (instruction rewriting, model upgrade) are neutral, confirming that the harm is specific to correction-injection patching.
We explain this asymmetry through the Linguistic Contract hypothesis: each downstream module implicitly adapts to its upstream's characteristic error distribution, so correcting the bottleneck breaks this implicit alignment in a way that upstream corrections do not. We operationalize this via a per-agent co-adaptation measure, derived from diagnosis alone, and show it is consistently associated with patching harm across agent families: higher co-adaptation co-occurs with harm, lower with safety. This trend holds across all three agent families, providing preliminary support for the hypothesis beyond a single-agent observation.
| Comments: | Preprint. Under review at EMNLP 2026 (ARR) |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.21958 [cs.CL] |
| (or arXiv:2605.21958v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.21958 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jeonghun Yoon [view email]
[v1]
Thu, 21 May 2026 03:44:47 UTC (76 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.