Selection Fronts at an Admissibility Boundary: Local Exit Geometry, Conditional Boundary Layers, and Reset Semantics

An admissibility boundary is the place where at least one declared control condition loses positive reserve. That static fact does not by itself produce a boundary layer, metastability, a large-deviation law, a measurement outcome, or a new physical phase. Those conclusions require a specified dynamical model. This paper gives a precise conditional meaning to the term selection front. The input is a finite vector of normalized signed reserve rows, including the rows needed for spectral separation, projector control, well-posedness, descent, coherent stability, truncation, and numerical certification. The exact boundary is distinguished from a conservative numerical event guard. At one regular active row the boundary is a smooth facet; at transverse row ties it is locally an orthant corner. A dynamical front additionally requires a drift, diffusion, initial law, stopping rule, interior return set, observables, and a post-exit stop, chart-transition, reset map, or reset kernel. The central calculation is an exactly solvable normal diffusion on an interval. Its guard-hitting committor and mean exit time are derived in closed form. Drift away from the guard gives an exponential boundary layer of thickness proportional to noise divided by drift; zero drift gives a linear committor and inverse-noise exit time; drift toward the guard gives neither behavior. Thus no universal sigmoid, metastable law, or first-passage scaling follows from admissibility alone. Eyring–Kramers and Freidlin–Wentzell asymptotics remain available when a small-noise family, attracting basin, quasipotential barrier, and regularity hypotheses are independently supplied. The result is a reproducible local theory of first exit and a strict application contract. It does not derive vacuum selection, black-hole dynamics, cosmological measures, quantum collapse, or limits on scalable coherence.