The same LLM runs twice on the same input. One side works from memory. The other has access to the Convergent Intelligence Engine — a structural knowledge layer that finds real connections across fields no chatbot or search engine can find alone.
◠ Working. This might take a minute or two.
◠ Working. This might take a minute or two.
Blind Evaluation
◠ Evaluating...
| Criterion | System A | System B + Underglow | Winner |
|---|
Both systems use the same base model and receive the same input document within the same prompt scaffold. The only difference is the final instruction: System A is asked to generate analogies from memory, while System B is asked to structure the input for retrieval against the Underglow Convergent Intelligence Engine. This is the minimum necessary divergence — without it, there would be nothing to compare. A blind judge scored both outputs without knowing which system had retrieval augmentation.