We assemble a large comprehensive sample of 2534 z~2, 3, 4, 5, 6, 7, 8, and 9 galaxies lensed by the six clusters from the Hubble Frontier Fields (HFF) program. Making use of the availability of multiple independent magnification models for each of the HFF clusters and alternatively treating one of the models as the "truth," we show that the median magnification factors from the v4 parametric models are typically reliable to values of 30-50, and in one case to 100. Using the median magnification factor from the latest v4 models, we estimate the UV luminosities of the 2534 lensed z~2-9 galaxies, finding sources as faint as -12.4mag at z~3 and -12.9mag at z~7. We explicitly demonstrate the power of the surface density-magnification relations {Sigma}(z) versus {mu} in the HFF clusters to constrain both distant galaxy properties and cluster lensing properties. Based on the {Sigma}(z) versus {mu} relations, we show that the median magnification estimates from existing public models must be reliable predictors of the true magnification {mu} to {mu}<15 (95% confidence). We also use the observed {Sigma}(z) versus {mu} relations to derive constraints on the evolution of the luminosity function faint-end slope from z~7 to z~2, showing that faint-end slope results can be consistent with blank-field studies if, and only if, the selection efficiency shows no strong dependence on the magnification factor {mu}. This can only be the case if very low-luminosity galaxies are very small, being unresolved in deep lensing probes.