Consulting Lab· Custom AI· AI Adoption· Bespoke Training· AI Welfare· Co-Author With Us
Field Guide · Research · Fascicle
published paper · applied apr mmxxvi

Metacognitive mode-switching reorganises cache geometry.

spectral entropy as a universal marker of metacognitive processing

Spectral entropy survives FWL correction as a universal marker across four models and three architectures. We measure the encoding-generation regime shift for the first time: when a model moves between consuming context and producing output, the cache silhouette reorganises in a stereotyped, model-independent way.

The shift is reliable enough to use as a state probe in real-time alignment systems and as a classifier feature in downstream geometry work.

Download PDF →

This research is ongoing. If your work intersects with ours —

Start a conversation → Research collaboration →