Automatic design of flight-test measurement chains requires assigning large numbers of sensors to limited acquisition resources under coupled engineering constraints. This paper proposes an executable-reward-guided configuration method that formulates sensor-to-board assignment as sequential interaction between a policy model and a deterministic configuration environment. The model observes a compact system state and generates structured actions that can be parsed, executed, diagnosed, and revised. A two-stage training procedure combines supervised fine-tuning with group-relative policy optimization. To align learning with engineering executability, the reward first rejects malformed or invalid actions and then evaluates constraint satisfaction, configuration efficiency, and final solution quality. This executable feedback guides the policy toward valid state transitions and resource-efficient configurations.

flight test,measurement chain,automatic configuration,executable reward,constraint-aware optimization