Machine-learning methods are increasingly applied to astronomical surveys, providing powerful tools for detecting and studying galaxy clusters. We investigate the mass-redshift properties and completeness of the ComPACT galaxy cluster catalogue, constructed using a convolutional neural network applied to publicly available combined ACT+Planck maps. The ComPACT catalogue contains 2962 SZ-selected galaxy cluster candidates. We confirm clusters by estimating redshifts using literature information and photometric techniques based on DESI Legacy Imaging Surveys data. Cluster masses are derived from ACT+Planck and Planck Compton-y maps via SZ scaling relations. The completeness is assessed using simulated cluster injections into real microwave maps. We confirm approximately ~60% of the ComPACT candidates as galaxy clusters. The redshifts span the range 0.007<z<1.7, including approximately 116 new measurements. Masses are obtained for 56% of the sample, covering the range (0.25-13.1)x10^14^M_{sun}_ and including 158 new mass determinations. We identify five previously unreported massive clusters (M_500c_>6x10^14~M_{sun}_) at z>0.7, increasing the known population of such systems by approximately 10%. The ComPACT catalogue expands the SZ-selected Planck-like cluster population, especially at high redshift and high mass, demonstrating the effectiveness of deep-learning approaches for cluster detection in microwave data.