Abstract: Designing the stacking sequence of a composite laminate is a discrete optimisation problem: each ply takes one of a few orientations, each evaluation may cost a finite-element solve, and the design must obey manufacturing rules and design guidelines, which optimisers enforce by penalty, repair or rejection. We describe the admissible set exactly. The two manufacturing rules forbid a finite set of factors, so the sequences they admit form a regular language, recognised by an automaton that tracks the last orientation and the length of its run; balance and the ten per cent rule constrain the Parikh vector and are carried by a dynamic program over automaton state and counters. One table, polynomial in N, counts the admissible sequences without producing them, draws them uniformly, and tells whether a partial sequence can still be completed. We give pseudocode, exact counts for three orientation sets, closed-form ceilings on the composition, and four of the five single-rule entries a published table left uncomputed. At 48 plies, 99.8% of the nodes of a layer-wise search tree checking the rules locally have no admissible completion on eight orientations; the completion test removes them. A one-point crossover that cuts two admissible parents only where they share an automaton state and their child closes the counters returns admissible children by construction.