What Oral Traditions Are About and What They Say

Cross-cultural databases are bounded by expert coding and missing observations. The Ethnographic Atlas (EA) codes 1,265 societies on 104 variables, with about 960 societies observed per variable on average, and the Standard Cross-Cultural Sample (SCCS) codes 186 societies on 1,717 variables, with about 105 observed per variable. I use a large language model and the 2,564-motif Berezkin catalogue to extract coverage, what motifs discuss, and stance, what positions they take. Stance, what a motif says, is strongest. Across 26 researcher-designed dimensions, stance is multidimensional (leading component 15.1%, matching ethnographic data) and largely independent of the coverage complexity gradient. On the two dimensions with direct human benchmarks, stance agrees on whether challenged characters succeed (kappa = 0.64) and whether antisocial behavior succeeds or is punished (kappa = 0.49), near the strongest available human benchmark of kappa about 0.57. Among 20 dimensions with no database equivalent, 18 have reliable group-level stance ratios across three runs (ICC >= 0.5$). Nine stance-outcome links survive continent fixed effects, false-discovery correction, and a complexity control, led by outsiders contact for lower GDP and emotional expression for higher female labor participation. Coverage succeeds differently. Across 1,805 usable codebook concepts, coverage predicts 65 of 88 EA variables and detects signal in 35 of 46 pooled SCCS domains, but this information concentrates in one factor. The first principal component explains 43% of score variance versus 15% in EA data, and the resulting complexity index predicts GDP within continents (R^2 = 0.55). Folklore measurement is one-dimensional only when the question is coverage. Researcher-defined concepts become measurable across 958 societies for $0.03 to $1.70.