Title: Superhorizon isocurvature fluctuations relax tensions

Abstract: We present a new class of models that have potential to alleviate tensions present in the cosmological data today. We postulate variation in the sound horizon scale on super-horizon scales, i.e. on scales that are larger than that of the present observable low-redshift universe ($\gtrsim 1\,$Gpc) while at the same time smaller than the largest scales probed by the cosmic microwave background (CMB) ($\lesssim10\,$Gpc). In this scenario, CMB peaks are naturally smoothed as preferred by the Planck data, while at the same time the low-redshift baryon acoustic oscillation calibration is partially decoupled from the CMB. Taking super-horizon variations in baryon fraction as an example and using approximate modeling, we find improvement in the best fit Planck power spectrum model $\Delta \chi^2 \sim 6$ for one extra degree of freedom with the relevant extension parameter $10^3 \sigma_b = 2.12 \pm 0.50 $, implying about 10\% variations in baryon fraction across the universe. At the same time, $S_8$ drops by about 1 sigma, easing tension with weak lensing surveys. While $H_0$ increases in this model by about 1 sigma, this is insufficient to explain the Hubble tension in $\Lambda$CDM. Since the power of low redshift BAO is relaxed, we find that the combination of Planck 2018 data, eBOSS BAO data and Riess et al distance ladder Hubble parameter determination produce a satisfactory fit in the model with free dark energy equation of state. Such a fit, however, favors a phantom dark energy equation of state $w<-1$ at 2-3 sigma.