Based on TESS observations of a δ Scuti variable WASP-33 obtained in 2019 and 2022, we thoroughly investigate the power spectrum of this target photometric flux and construct a statistically exhaustive model for its variability. This model contains 30 robustly justified harmonics detected in both the TESS sectors simultaneously, 13 less robust harmonics detected in a single TESS sector, a red noise, and a quasi-periodic noise term. This allowed us to greatly improve the accuracy of the exoplanet WASP-33 b transit timings, reducing the TTV residuals r.m.s. drastically, by a factor of 3.5, from 63 to 18 s. Finally, our analysis does not confirm existence of a detectable orbital phase variation claimed previously based on WASP-33 TESS photometry of 2019.