Abstract
We assess the robustness of $\Lambda$CDM results from the full-shape analysis of BOSS power spectrum using the one-loop prediction of the Effective Field Theory of Large-Scale Structure (EFTofLSS). The public likelihoods PyBird and CLASS-PT lead to results in agreement only at the $1\sigma$ level, despite the fact that they are derived from the same BOSS dataset and theory model. We perform a thorough comparison of the various analyses choices made between the two pipelines, and identify that the differences come from the choice of prior on the EFT parameters, dubbed "West-coast" (WC) and "East-coast" (EC) prior, respectively associated to PyBird and CLASS-PT. In particular, because posteriors are non-Gaussian, projection effects from the marginalization over the EFT parameters shift the posterior mean of the cosmological parameters with respect to the best-fit up to $1\sigma$ in the WC prior and up to $2\sigma$ in the EC prior. We quantify that best-fit cosmological parameters extracted from BOSS given the two prior choices are consistent at $\sim 1\sigma$. The consistency improves to $\sim 0.5\sigma$ when doubling the prior widths. While this reveals that current EFT analyses are subject to prior effects, we show that cosmological results obtained in combination with CMB, or from forthcoming large-volume data, are less sensitive to those effects. In addition, we investigate differences between BOSS measurements. We find broad agreements across all pre-reconstructed measurements considered ($<0.6\sigma$), but the two available BOSS post-reconstructed measurements in Fourier space, once combined with the EFT full-shape analysis, lead to discrepant Hubble parameter $H_0$ at $\sim 0.9\sigma$. Given the various effects we discuss, we argue that the clustering amplitude $\sigma_8$ measured with BOSS is not in statistical tension with that inferred from Planck under $\Lambda$CDM.