Cosmological tensions and systematic cross-checks

The Hubble rate inferred from microwave-background spectra is a derived parameter linking early-universe physical scales and angular geometry through an assumed cosmology. A local distance ladder instead links geometric anchors, stellar indicators, and standardized supernovae to nearby expansion. Both can estimate the present rate, but their calibration and model dependencies differ. Neither description means one route has no assumptions. The comparison must specify background geometry, early-time physics, data selections, and uncertainty conventions.

For a historical exercise, the Planck 2018 base flat Lambda cold-dark-matter analysis of temperature, polarization, and lensing reports H zero = 67.4 plus or minus 0.5 km per second per megaparsec. The SH0ES 2022 baseline Cepheid–supernova analysis reports 73.04 plus or minus 1.04 in the same units. These are explicitly dated primary results, not a claim about the latest consensus or every analysis variant. Their one-number summaries permit an instructive calculation while omitting information contained in full likelihoods and alternative calibrations.

Shared calibration can correlate estimates. A common error with the same sign may cancel in their difference, so positive covariance reduces the difference variance; adding all common errors independently can actually understate that particular discrepancy. Other shared ingredients can create more complicated dependence, including data reused in an inverse distance ladder. One must derive the covariance from the measurement model instead of assuming that every systematic broadens every comparison in the same way.

Multidimensional consistency cannot generally be summarized by selecting the most discrepant one-dimensional parameter after examining many projections. A covariance-weighted difference tests a specified vector under a Gaussian approximation, while posterior-predictive checks test replicated observables and Bayesian evidence asks another question. Priors, parameter boundaries, non-Gaussian likelihoods, and scanning many models can invalidate a naive sigma interpretation. The statistic and the question it answers should travel together whenever a tension is quoted.

A useful cross-check changes a vulnerable dependency while preserving enough shared information to diagnose it. Distance ladders can compare geometric anchors, stellar populations, crowding corrections, passbands, and supernova host matching. Independent distance methods introduce different astrophysical calibrations but still require their own selection and likelihood models. Microwave analyses can vary frequency, sky region, foreground treatment, polarization, and experiment. Internal splits are valuable even when they are correlated; their interpretation must use the appropriate split covariance.

A proposed early-time change that reduces the sound horizon may alter an inferred Hubble rate while also shifting acoustic phases, damping, abundances, or growth. A late-time expansion change must confront distance and clustering observations together. Matching one scalar summary is therefore not sufficient evidence for a complete explanation. Blinded choices, injection tests, reproducible likelihoods, and genuinely different observations help identify whether a discrepancy survives. We do not assert a resolution or an up-to-date significance; this lesson demonstrates a dated comparison and a method for evaluating future evidence.