known_biases: Individual-participant pooling of 6 heterogeneous US cohorts (ARIC, CARDIA, CHS, Framingham Offspring, Jackson Heart, MESA) with mostly a single baseline diet assessment harmonized across different FFQ/diet-history instruments — shared vulnerability to regression-dilution from one-time diet measurement, cross-instrument harmonization error, and residual confounding by baseline lifestyle (the “healthy/unhealthy eater” cluster) that moves every finding built on it together.