Spontaneous photon bursts are observed in the output collected from a mesoscale semiconductor-based laser near the lasing threshold. Their appearence is compared to predictions obtained from Laser Rate Equations and from a Stochastic Laser Simulator. While the latter is capable of predicting the observed large photon bursts, the photon numbers computed by the former produces a noisy trace well below the experimentally detectable limit. We explain the discrepancy between the two approaches on the basis of an incorrect accounting of the onset of stimulated emission by the Rate Equations, which instead are capable of complementing the physical description through topological considerations.