Diagnosis Is Not Prescription: Linguistic Co-Adaptation Explains Patching Hazards in LLM Pipelines

TL;DR

The paper reveals that in multi-module LLM agents, fixing the routing module often harms performance due to linguistic co-adaptation, while upstream corrections tend to improve outcomes, highlighting the importance of understanding module interactions.

Contribution

It introduces the Diagnostic Paradox in LLM pipelines and the Linguistic Contract hypothesis, explaining why certain patches degrade performance and how co-adaptation measures predict patching harm.

Findings

Routing module diagnosis identifies the bottleneck but patching it harms performance.

Upstream query-rewriting modules, when patched, reliably improve outcomes.

Higher co-adaptation correlates with patching harm across multiple agent families.

Abstract

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.