12 Repairing Kan-Extension Structure

LINCS–KET

Kan Extension Transformers provide a clean architectural test of LINCS. The model contains a direct route and an incidence-restricted Kan-extension route to a common representation. Their intended agreement is structural. A post-training repair may reduce the route obstruction, but it is admitted only while the shared predictive task remains intact.

¹1. This chapter uses KET as the architecture and the later registered LINCS studies as the evidence. The original KET experiments are not retroactively counted as LINCS experiments. ↩