Universal and non-universal signatures in the scaling functions of critical variables

The view that the probability density function (PDF) of a key statistical variable, anomalously scaled by size or time, could furnish a hallmark of universal behavior contrasts with the circumstance that such density sensibly depends on non-universal features. We solve this apparent contradiction by demonstrating that both non-universal amplitudes and universal exponents of leading critical singularities in large deviation functions are determined by the PDF tails, whose form is argued on extensivity. This unexplored scenario implies a universal form of central limit theorem at criticality and is confirmed by exact calculations for mean field Ising models in equilibrium and for anomalous diffusion models.