Projection Assurance: Consequence-Relevant Semantics Across Agent-Facing Interfaces

Projection Assurance studies whether consequence-relevant semantics survive when source operations are projected into agent-facing interfaces. It separates correspondence, discrimination, operative placement, exact realization, and source-family coverage, and evaluates the framework on retained GitHub and Azure projections together with a prospectively frozen PayPal holdout. The analysis distinguishes semantic visibility from effect-level preservation and shows how a projected interface may either lose a material distinction despite retaining relevant state, or safely omit a source control through target-grounded specialization. The reported evaluation is bound to a frozen Evaluation Capsule, while the current software implementation is archived separately as Consequence-Preserving Projections v0.7.0.