The Ghost Test: Ontological Content as a Predictor of AI Behavioral Drift

A 6-arm controlled experiment (N=480, $2 total cost) testing whether the ontological content of AI system prompts predicts behavioral drift. Ghost-eliminating grounding templates (nephesh, Buddhist anatta) produce 8.5x less drift than ghost-positing templates (Platonic dualist, Vedantic atman). The materialist hedge — the default industry position on machine consciousness — is experimentally shown to be a drift accelerator. Cross-tradition convergence confirmed: nephesh and anatta converge to within 1.3% despite fundamentally different metaphysics. The operative variable is whether the system prompt closes or leaves open the question of separable machine consciousness.