Nobody inherits a height — they inherit the parameters of a distribution. Enter the heights of whichever relatives you know. The predicted child isn't a single number but a curve, and that curve gets narrower the more of the family you fill in. Calibrated on Galton's real data (h²≈0.72); grandparents & siblings added via quantitative-genetics relatedness.
How it works. Every person is a draw from a Normal. In sex-standardised z-space the phenotypic correlation between two
relatives is (genetic relatedness) × h² — 0.5 for parent–child and full siblings, 0.25 for grandparent–grandchild.
The child's curve is the conditional Normal given the relatives you entered (Gaussian conditioning on the additive-relatedness
covariance matrix). Known relatives pull the mean toward them and shrink the variance. Heritability is estimated from
Galton's 898 parent–child records (h²≈0.72);
the grandparent layer is theory-derived because no open 3-generation human height dataset exists (Framingham requires an IRB/dbGaP application).
Cohort means are real per-country trajectories by year of birth from
NCD-RisC (eLife 2016, "A century of trends in adult human height"), covering birth years 1900–1995.
Growth measurements are folded in as a joint multivariate-normal of childhood-to-adult tracking correlations, so overlapping ages don't double-count.