Fatigue is a difficult multi-scale modelling problem nucleating from localised plasticity at the scale of dislocations and microstructure with significant engineering safety implications. Cold dwell fatigue is a phenomenon in titanium where stress holds at moderate temperatures lead to substantial reductions in cyclic life, and has been implicated in service failures. Using discrete dislocation plasticity modelling complemented by transmission electron microscopy, we successfully predict lifetimes for 'worst case' microstructures representative of jet engine spin tests. Fatigue loading above a threshold stress is found to produce slip in soft grains, leading to strong dislocation pile-ups at boundaries with hard grains. Pile-up stresses generated are high enough to nucleate hard grain basal dislocations, as observed experimentally. Reduction of applied cyclic load alongside a temperature excursion during the cycle lead to much lower densities of prism dislocations in soft grains and, sometimes, the elimination of basal dislocations in hard grains altogether.