Finite-Library Identity-Label Auditing under Partial Observation

This technical report presents a reproducible audit procedure for a fixed finite library of discrete-time autonomous linear systems under partial observation. Candidate realizations carry pre-certified identity labels relative to a reference system. Given a known observation map, an unknown initial state and a declared bounded observation error, linear programming retains candidates consistent with the observations. Unanimous surviving labels permit a decision; conflicting labels require abstention. Empty feasible sets and numerical uncertainty are reported separately. The reproducibility package contains eleven synthetic cases covering equivalent realizations, same-spectrum negative controls, hidden-state ambiguity, observation-information changes and error-bound boundaries. All eleven expected outcomes were reproduced. Verification additionally checked 33 exact rational linear-programming optimality certificates, 27 aggregation states and three malformed-solver probes. The procedure is an engineering specialization of established set-membership filtering and selective agreement principles. Its guarantees depend on candidate-library coverage and a valid error bound; it does not establish general dynamical identity from finite observations.