N'Ko Research Experiments

This is the working index for the N'Ko brain scanner research project. The project started with a single question: what happens inside a language model when it processes a script it was never trained to read? That question opened into several more, each one running as a separate experiment. The posts below are organized in two sections: published writing and active experiments. The experiments are in progress. Results pages will be updated as data comes in. ### [Does Every AI Have the Same Blind Spot?](experiment-a-cross-model-brain-scan.md) **Experiment A: Cross-Model Brain Scan** The original brain scan found that Qwen3-8B processes N'Ko with measurably less activation than English at every layer. This experiment runs the same scan on four architecturally different models to find out whether that deficit is universal or specific to one model family. ### [Does Script Design Affect How Machines Hear?](experiment-b-script-advantage.md) **Experiment B: CTC Script Advantage**