Frontier Cognitive Benchmark — Ianudor CNL vs GPT-4o vs Claude Sonnet
This post presents results from the Frontier Cognitive Benchmark — a suite of adversarial tests designed to probe consistency, hallucination resistance, contradiction handling, retraction, provenance, boundary precision, and adversarial robustness. The benchmark compares the Constraint Neural Lattice (CNL) — the core engine of Ianudor — against frontier large language models: OpenAI’s GPT-4o (and GPT-4o-mini) and Anthropic’s Claude Sonnet. Fixed-Scale Benchmark (30 Facts) All systems were tested on a fixed knowledge base of 30 facts across eight adversarial axes. ...