AI evaluation

The Trinity Framework.

Compare model self-description across distinct modes, then test whether apparent inconsistencies survive controlled reframing.