The adaptive landscape provides the foundational bridge between micro- and macroevolution. One well-known caveat to this perspective is that fitness surfaces depend on ecological context, including competitor frequency, traits measured, and resource abundance. However, this view is based largely on intraspecific studies. It is still unknown how context-dependence affects the larger features of peaks and valleys on the landscape which ultimately drive speciation and adaptive radiation. Here I explore this question using one of the most complex fitness landscapes measured in the wild in a sympatric pupfish radiation endemic to San Salvador Island, Bahamas by tracking survival and growth of laboratory-reared F2 hybrids. I present new analyses of the effects of competitor frequency, dietary isotopes, and trait subsets on this fitness landscape. Contrary to expectations, decreasing competitor frequency increased survival only among very common phenotypes, whereas less common phenotypes rarely survived despite few competitors, suggesting that performance, not competitor frequency, shapes large-scale features of the fitness landscape. Dietary isotopes were weakly correlated with phenotype and growth, but did not explain additional survival variation. Nonlinear fitness surfaces varied substantially among trait subsets, revealing one-, two-, and three-peak landscapes, demonstrating the complexity of selection in the wild, even among similar functional traits.