Matthieu Verstraete Université de Liège Thermoelectricity has undergone a strong revival in the past 20 years. First theoretical predictions then lab measurements have shown that materials with high figure of merit can be produced. Industrial scaling up and a killer compound still need to be found, whether for room-temperature applications or high power high temperature waste heat recovery in thermal engines. The importance of numerical simulations in the field has been extreme, first in showing that nanostructuring can break certain scaling laws between transport coefficients, and now in allowing high-throughput scans of materials, defects, and doping levels. I will present evidence that experimentalists should be enthusiastic about but wary of theoretical predictions - simplistic assumptions about the electronic structure or the behavior of scattering mechanisms can lead to quantitatively but also qualitatively incorrect results. In particular I will show how the single-parabolic-band model has made Hicks and Dresselhaus' seminal scaling relations almost impossible to realize[1,2], and how the constant relaxation time approximation should break down much more often than it does[3,4]. [1] Filippetti, A. et al. Phys Rev B 86, 195301 (2012) [2] Delugas et al. Phys. Rev. B 88, 045310 (2013) [3] Xu and Verstraete Phys. Rev. B 87, 134302 (2013) [4] Xu and Verstraete Unpublished