Treatment Effects from Unit Beliefs: Orthogonal Estimation Beyond Covariate Heterogeneity

USL-01 type contract

The Unit Selection Variable $U\sim\Pi$ performs individual selection under the task-declared Population law $\Pi$. Pre-treatment evidence $\mathcal O^-$ first conditions the world law $P(U\in du\mid\mathcal O^-)$; the learner forms $Q_\phi(du\mid\mathcal O^-)\approx P(U\in du\mid\mathcal O^-)$ and must evaluate that approximation separately. Within the task-declared unit space, $u$ is the formal Individual and selects the whole treatment-response-law member, such as $P_\theta(dy\mid a,x,u,\mathcal O^-)$. This is a task-level modeling convention, not a metaphysical claim. The response interface is distinct from attribution, but the same $u$ indexes it directly; no second foundational representation is required. An application may optionally derive a property map $\rho:u\mapsto Z_u$ and use the pushforward $R_\phi(dz\mid\mathcal O^-)=\rho_{\#}Q_\phi(dz\mid\mathcal O^-)$, or define a distance between fixed-individual response laws. Neither $Q_\phi$ nor the measurable selector domain $\mathcal U$ supplies response geometry by itself. A response map may omit $\mathcal O^-$ only under a declared response-sufficiency assumption. All evidence in $\mathcal O^-$ must be pre-treatment, and this type split does not supply causal identification without explicit assignment, overlap, consistency and nuisance-function assumptions. The world causal estimand must be fixed before introducing $Q_\phi$; replacing the world conditional by a learner belief creates an estimator or approximation, not a new causal target unless a learner-relative decision target is explicitly declared.