No single party holds the shadow participant: the state knows one thing about you, a physician another, an employer another, your friends far more. What makes it qualitatively different — and dangerous — is integration. An AI that combines those streams produces a single externalized model of the person, and the deep failure the framework warns about is that institutions may come to prefer interacting with the model because it is easier than encountering the actual human being.
This is the link between surveillance and gatekeeper AI: surveillance assembles the shadow participant, and the gates then judge the person against it — so winning one appeal changes little while the underlying model keeps circulating. It is the behavioral cousin of the universal participant score, and the framework resists it directly through anti-persistence architecture, a ban on universal behavioral dossiers, and firebreaks that stop every local observation from becoming universal.