It has been hypothesized that even ‘perfect’ polygenic scores (PGSs) composed of only causal variants may not be fully portable between different social groups owing to gene-by-environment interactions modifying the expression of relevant variants. The impacts of such interactions involving two forms of social adversity (low socioeconomic status [SES] and discrimination) are examined in relation to the expressivity of a PGS for educational attainment composed of putatively causal variants in a large, representatively sampled and genotyped cohort of US children. A relatively small-magnitude Scarr–Rowe effect is present (SES × PGSEDU predicting General Cognitive Ability [GCA]; sR = .02, 95% CI [.00, .04]), as is a distinct discrimination × PGSEDU interaction predicting GCA (sR = −.02, 95% CI [−.05, 00]). Both are independent of the confounding main effects of 10 ancestral principal components, PGSEDU, SES, discrimination and interactions among these factors. No sex differences were found. These interactions were examined in relation to phenotypic and genotypic data on height, a prospectively more socially neutral trait. They were absent in both cases. The discrimination × PGSEDU interaction is a co-moderator of the differences posited in modern versions of Spearman’s hypothesis (along with shared environmentality), lending support to certain environmental explanations of those differences. Behavior-genetic analysis of self-reported discrimination indicates that it is nonsignificantly heritable (h2 = .027, 95% CI [−.05, .10]), meaning that it is not merely proxying some underlying source of heritable phenotypic variability. This suggests that experiences of discrimination might stem instead from the action of purely social forces.