The Physically Admissible System Cut - When a tensor factorization is a physical subsystem: an exact and a quantitative criterion

A tensor factorization specifies a mathematical subsystem, but it does not by itself establish that its reduced state is a complete description of the physical bearer it claims to represent. This paper supplies an exact selection criterion and a quantitative certificate without altering the global physics. It retains the observable-induced algebraic formulation, the quotient-dynamics theorem, the strict two-qubit counterexample, the autonomy defect, additive error accumulation, nested persistence margins, the coupling–time estimate and the non-circular selection argument. The physical-cut theorem now distinguishes three independent descent requirements: complete constitution, retained existence and identity verdicts, and autonomous dynamics. A realization x=(c,ρ,η) separates the physical profile, quantum state and preparation/history/continuation data. This prevents a historical identity verdict from being silently treated as a function of a momentary density matrix. The CNOT-conjugated cut has identical reduced inputs and orthogonal reduced outputs at t=π/4. A second witness has perfectly closed baseline dynamics while a partial trace erases two physically distinguishable coupling profiles. Dynamic closure alone therefore does not establish physical completeness. The quantitative results bound unavoidable prediction error from below by half the fibre defect and, with an admissible lift in the chosen model class, from above by that defect. Reachable-state error bounds accumulate under CPTP comparison. A bound below an independently grounded Q1, Q2a or Q2b margin is a sufficient persistence certificate; loss of that inequality is not an identity-loss verdict. For a bounded interaction the specified decoupled predictor has the sharper error bound |λ|(t−s)‖V‖/ℏ, while the fibre defect has twice that bound. The resulting criterion links the physical cut to emergence, thermodynamic process closure and record memory without identifying those questions. All exact theorems below are finite-dimensional and relative to explicitly named preparation and continuation domains.