Affordance-Compiled Intelligence: Observable-Only Cognitive Impedance Matching for No-Meta LLM-Integrated Systems
Patrick Lewis · 2026 · 2988 citationsRead the paper
Affordance-Compiled Intelligence develops Cognitive Impedance Matching Theory (CIMT), an observable-only and no-meta protected compiler theory for LLM-integrated systems. The paper studies how a fixed model-policy can exhibit different operational capability when the surrounding world is redesigned through observations, typed action handles, validators, repair paths, rollback modes, authority scopes, context summaries, and auditable receipts. CIMT treats system-level capability amplification as a world-side compilation problem rather than a model-weight improvement problem. It defines operational claims through explicit claim objects and evidence objects, using committed 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. The theory 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. It also includes worked examples for code-editing agents and retrieval-augmented generation systems. The intended contribution is a practical formal foundation for making fixed-model LLM systems more reliable through observable world-side interface, authority, validation, repair, and audit design.
What the full text shows
Working prototype- RAG-Sequence achieved state-of-the-art results on three open-domain QA tasks (NQ, WQ, CT). — NQ: 44.5 EM, WQ: 45.2 EM, CT: 52.2 EM (Compared against closed-book and open-book baselines)
- RAG can generate correct answers even when the answer is not present in any retrieved document. — 11.8% accuracy vs 0% for extractive models (Natural Questions (NQ) dataset)
- RAG-Sequence outperformed BART on Open MS-MARCO NLG. — 2.6 Bleu points and 2.6 Rouge-L points higher (Abstractive Question Answering)
- RAG-Token outperformed BART in factual accuracy for Jeopardy question generation. — RAG more factual in 42.7% of cases vs BART more factual in 7.1% of cases (Human evaluation of 452 generation pairs)
- RAG-Token outperformed BART on Jeopardy question generation. — Higher Q-BLEU-1 score (Jeopardy Question Generation task)
- Setup
- Wikipedia December 2018 dump (21M 100-word chunks) indexed via FAISS HNSW; BART-large (400M parameters) as generator; datasets include NQ, TQA, WQ, CT, MSMARCO, and SearchQA.
- How
- The researchers implemented Retrieval-Augmented Generation (RAG) by combining a DPR bi-encoder retriever with a BART-large seq2seq generator. They tested two marginalization strategies: RAG-Sequence (one document for the whole sequence) and RAG-Token (different documents per token).
- Limitations
- Performance on MSMARCO is limited by the absence of gold passages and some questions being unanswerable via Wikipedia alone. · Potential for bias or inaccuracies inherited from the Wikipedia source. · Potential for misuse in generating misleading content or spam.
Summarized by Gemma 4 from the open-access full text · 2 Oct 2026. Read it