The Clarity Challenge provides an excellent opportunity to stimulate novel, performant machine learning approaches to hearing aid signal enhancement. It is important that these methods are compared with classical methods which are well understood. Here, an adaptive beamformer based on the minimumvariance distortionless response design approach is proposed as a superior baseline against which machine learning approaches can be benchmarked. The design exploits documented characteristics of the Challenge rules to identify noise-only segments and the direction-of-arrival of the target. Hearing-aid specific modifications include automatic gain control and listenerspecific hearing loss compensation. On the dev dataset the proposed method obtains a mean MBSTOI metric of 0.61 compared to the baseline system which achieves 0.41.