Big bang nucleosynthesis provides the earliest probe of standard model physics, at a time when the universe was less than a thousand seconds old. It determines the abundances of the lightest nuclides, which give rise to the subsequent history of the visible matter in the Universe. This work derives new $^7$Be(n,p)$^7$Li thermonuclear reaction rates based on all available experimental information. This reaction sensitively impacts the primordial abundances of $^{7}$Be and $^7$Li during big bang nucleosynthesis. We critically evaluate all available data and disregard experimental results that are questionable. For the nuclear model, we adopt an incoherent sum of single-level, two-channel R-matrix approximation expressions, which are implemented into a hierarchical Bayesian model, to analyze the remaining six data sets we deem most reliable. In the fitting of the data, we consistently model all known sources of uncertainty, including discrepant absolute normalizations of different data sets, and also take the variation of the neutron and proton channel radii into account, hence providing less biased estimates of the $^7$Be(n,p)$^7$Li thermonuclear rates. From the resulting posteriors, we extract R-matrix parameters ($E_r$, $\gamma^2_n$, $\gamma^2_p$) and derive excitation energies, partial and total widths. Our fit is sensitive to the contributions of the first three levels above the neutron threshold. Reaction rates were computed by integrating 10,000 samples of the reduced cross section. Our $^7$Be(n,p)$^7$Li thermonuclear rates have uncertainties between 1.5% and 2.0% at temperatures of $\leq$1 GK. We compare our rates to previous results and find that the $^7$Be(n,p)$^7$Li rates most commonly used in big bang simulations have too optimistic uncertainties.