Correction and Corruption: A Two-Rate View of Error Flow in LLM Protocols

arXiv:2604.18245v1 Announce Type: new Abstract: Large language models are increasingly deployed as protocols: structured multi-call procedures that spend additional computation to transform a baseline answer into a final one. These protocols are evaluated only by end-to-end accuracy, giving limited insight into when they help, when they hurt, and whether their behavior transfers under distribution shift or composition. We propose a paired-outcome measurement interface for auditing a single protocol step on exact-match tasks. For each instance, the interface records a baseline correctness bit E_0in{0,1} and a post-step correctness bit E_1in{0,1}, separating correction (E_0=0o E_1=1) from corruption (E_0=1o E_1=0) through two rates: c=Pr(E_1=1mid E_0=0) and gamma=Pr(E_1=0mid E_0=1). These rates predict accuracy changes and define a reusable empirical interface testable across seeds, mixtures, and pipelines. We identify three failure mechanisms. Under mixture shift, pooled estimates of (c,gamma) become biased when calibration and deployment mixtures differ; conditioning on a difficulty proxy restores stability without additional model calls. Under presentation contamination, selection protocols alter the interface through stable presentation artifacts when candidate content is fixed. Under state insufficiency, the correctness bit may not carry enough history for multi-step pipelines to compose predictably; a Markov factorization test identifies when composition is valid and where additional state is needed. When a protocol step passes these diagnostics, it becomes an auditable module: gated by estimated gain, conditioned on a difficulty proxy to correct mixture bias, and composed into multi-step pipelines with predictable accuracy. We demonstrate these ideas on synthetic mathematical tasks and on GSM8K, where the calibrated interface correctly predicts when protocol steps should be activated or suppressed.

Related

• Robust Tool Use via Fission-GRPO: Learning to Recover from Execution Errors
• Towards Robust Endogenous Reasoning: Unifying Drift Adaptation in Non-Stationary Tuning
• The Illusion of Certainty: Decoupling Capability and Calibration in On-Policy Distillation
• Do Not Step Into the Same River Twice: Learning to Reason from Trial and Error
• ORBIT: On-policy Exploration-Exploitation for Controllable Multi-Budget Reasoning
• Enhancing LLM-based Search Agents via Contribution Weighted Group Relative Policy Optimization

Source: arXiv cs.LG | 2026-04-21