Counterfactual Self-Location in Artificial Models

This record contains the manuscript Counterfactual Self-Location in Artificial Models, which develops a formal framework for model self-location. A model's own input, processing, and output histories become objects of counterfactual perturbation. Three operators — input (exogenous), recursion (interventional), and output (counterfactual) — form a causal hierarchy, and their pairwise interactions generate a 3×3 sensitivity matrix identified with the Jacobian of the model with respect to itself. Eigendecomposition of the symmetric part yields a self-sensitivity structure, and a fourth axis, an invariant functional, distinguishes genuine self-knowledge from optimization-induced artifacts. The manuscript extends the framework through dynamics, reflexivity, multiplicity, multi-model interaction, grounding, an action layer, personality as residual structure, and neural grounding, and it identifies the grounding of the invariant as the principal open problem.