We provide a comprehensive theoretical framework and a quantitative test of the method we recently proposed for processing data from a spherical detector with five or six transducers. Our algorithm is a trigger event generator performing a coherent analysis of the sphere channels. In order to test our pipeline we first built a detailed numerical model of the detector, including deviations from the ideal case such as quadrupole modes splitting, and non-identical transducer readout chains. This model, coupled with a Gaussian noise generator, has then been used to produce six time series, corresponding to the outputs of the six transducers attached to the sphere. We finally injected gravitational wave burst signals into the data stream, as well as bursts of non-gravitational origin in order to mimic the presence of non-Gaussian noise, and then processed the mock data. We report quantitative results for the detection efficiency versus the false alarm rate and for the affordability of the reconstruction of the direction of arrival. In particular, the combination of the two direction reconstruction methods can reduce by a factor of 10 the number of false alarms due to the non-Gaussian noise.