Abstract
We investigate the usefulness of side entanglement in discriminating between two generic qubit channels, {\ up to unitary pre- and post-processing,} and determine exact conditions under which it does enhance (as well as conditions under which it does not) the success probability. This is done in a constructive way by first analyzing the problem for channels that are extremal in the set of completely positive and trace-preserving qubit linear maps and then for channels that are inside such a set.