Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis et al.
We propose Cognitive Impedance Matching Theory (CIMT), a compiler theory that enhances the capabilities of a fixed LLM model through world-side interface, validation, repair, and audit design.
The reliability of LLM-integrated systems depends not only on model improvements but also on the design of the surrounding world (interfaces, authorities, validations, repairs, audits). However, existing approaches focus on model weight improvements or treat human/LLM evaluators as absolute standards, making systematic assurance difficult. There is a need for a theoretical foundation that can improve system-level capabilities through world design even with a fixed model policy.
CIMT treats system-level capability amplification as a world-side compilation problem rather than a model-weight improvement problem. It defines observable ledgers, target-evaluation channels, deterministic reducers, validity budget ledgers, evidence dependency graphs, artifact I/O manifests, conformance envelopes, and finite-sample or sequential certificates. Human reviewers, LLM judges, benchmarks, and external auditors are not treated as privileged evaluators; they are modeled as named, fallible measurement channels. This provides a conservative certification framework for paired target-channel improvement, vector debt accounting, forbidden-coordinate zero certificates, target-firewall discipline, scope simulation, dynamic widening, runtime and model-policy conformance, macro reliability, repair contraction, distribution-shift transfer, and receipt sufficiency.
Worked examples for code-editing agents and retrieval-augmented generation systems are presented, providing a practical formal foundation for making fixed-model LLM systems more reliable. Unlike existing model-centric approaches, this work systematizes world-side design, establishing a theoretical basis to improve system reliability without model updates.