Forward simulations of deleterious and adaptive variation. a) Alternative simulated demographic trajectories. The 1X trend represents the known ancestral size of the Seychelles paradise flycatcher based on the reconstruction of the recent demography with GONE (Fig. 2c). The alternative scenarios represent medium (5X) and large (10X) ancestral population sizes. The trajectory was divided into six stages: Ancestral (years 1810–1815), Collapse (1965–1970), Recovery (1990–1995), Present (2010–2015), Environmental change (2050–2055), Future (2095–2100). During the environmental shift, the quantitative trait optimum value moved from 0.2 to 1.2, resulting in a loss of fitness followed by adaptive evolutionary change. b) Genetic load. c) Masked load. d) Realized load. We calculated the genetic load components, following Bertorelle et al. (2022): $Geneticload=∑i=1Lqisi$⁠, $Realisedload=∑i=1Lqi2si+2∑i=1Lqi[1−qi]hisi$⁠, and masked load = genetic load−realized load. Furthermore, s_i is the selection coefficient, h_i is the dominance coefficient, and q_i is the frequency of the mutation at loci L. The genetic load, masked load, and realized load are all in lethal equivalents (see Bertorelle et al. 2022). The reduction in fitness (w) due to the expression of unconditionally deleterious mutations (i.e. inbreeding depression) is a function of the realized load: $w=e−RealisedLoad$⁠. e) Extinction probability per scenario (the number of surviving replicates divided by the total number of replicates). f) Additive genetic variance in the quantitative trait. g) Fitness effect conferred by the quantitative trait.