Beschreibung
As cosmological datasets from Large-Scale Structure (LSS) and the Cosmic Microwave Background (CMB) grow in precision and complexity, disparities between datasets demand robust statistical methods for model comparison and tension quantification. However, calculating Bayesian evidences remains a computationally prohibitive bottleneck, limiting the rapid validation of new datasets and extended models. To address this challenge, I will introduce unimpeded [arXiv:2511.04661, 2511.05470]. Conceptually analogous to the Planck Legacy Archive (PLA), which provides MCMC chains restricted to parameter estimation without direct access to Bayesian evidences, unimpeded is a reusable library of nested sampling chains and Bayesian evidences, enhanced by machine learning emulators utilising normalising flows. Spanning a massive pre-computed grid across 13 cosmological surveys and 8 models, it features built-in tools for systematic tension quantification. unimpeded acts as a scalable inference tool, transforming months of supercomputer time into seconds on a laptop.
To demonstrate unimpeded's ability to untangle complex inter-dataset systematics, I will present a fully Bayesian reanalysis of the recent DESI DR2 BAO data [arXiv:2511.10631, 2603.05472]. I will show that the Bayesian evidence modestly favours ΛCDM (ln B = -0.57 ± 0.26), in contrast to the Collaboration's 3.1σ frequentist preference for w0waCDM. I will explain how this discrepancy between Bayesian and frequentist conclusions arises from the Jeffreys-Lindley paradox. When extending this analysis to include the DES-SN5YR supernova catalogue, our Bayesian analysis yields a 3.07 ± 0.10σ preference for w0waCDM, notably lower than the Collaboration's reported 4.2σ frequentist significance. By employing the unimpeded tension metrics, I will pinpoint the origin of this preference: a significant (2.95 ± 0.04σ), low-dimensional tension between DESI DR2 and DES-SN5YR that is present only within the ΛCDM framework. The w0waCDM model is preferred precisely because its additional parameters act to resolve this specific dataset conflict. Notably, once the recalibrated DES-Dovekie supernova calibration is applied, this tension is mitigated and the Bayesian evidence for dynamical dark energy vanishes.