An arXiv preprint reports a 5.4% mismatch between the two main SDSS transverse BAO compilations. A direct comparison put the difference at (5.4 ± 1.4)%, or 3.9σ, and found that the offset was constant across redshift.
The paper's central warning is that a constant calibration error can be invisible when analysts use transverse BAO alone. In that case, an error in θBAO is exactly degenerate with a shift in rd h, so the fit quality and the plausibility of the fitted parameters can remain unchanged.
A cross-check designed to expose a hidden offset
The proposed check uses radial BAO as a reference for transverse measurements through a flat-FLRW consistency relation. In its normalized form, the relation cancels the ruler scale rd and contains neither H0 nor w(z).
The integrated version fits the relation as a straight line. Leaving the intercept and slope free provides an anchor-free linearity test; an independent distance anchor converts the slope into ε, a measure of relative transverse calibration.
The data behind the comparison
The radial reference combined BOSS DR12 consensus measurements at redshifts 0.38, 0.51 and 0.61 with DESI DR2 measurements above 0.61. The anchored BOSS distance at z0 = 0.38 was DM/rd = 10.231 ± 0.166.
The transverse inputs were labelled MM and N20. MM supplied 14 measurements from z = 0.35 to 0.63 and a 14 × 14 correlation matrix; N20 supplied 12 measurements from z = 0.365 to 0.65 and did not apply an equivalent correction.
The mismatch emerges after calibration
Before the external anchor was applied, both data sets passed the straight-line check. N20 gave χ2/dof = 11.6/10 with p = 0.31, and MM gave 11.2/12 with p = 0.52. The analysis found no evidence of a departure from the flat-FLRW redshift relation.
Once the anchor was used, N20 returned ε = 1.073 ± 0.021, while MM returned ε = 1.021 ± 0.029. With ε = 1 as the no-offset benchmark, the N20 result differed from unity at 3.8σ, while MM was consistent with unity.
The error bars tell a second story
The N20 significance carries an important caveat. When the analysis varied an assumed common correlation from ρ = 0 to 0.5, the central ε changed by less than 0.001. But at ρ = 0.3, the significance fell from 3.8σ for independent points to 2.7σ. The common angular covariance was not published.
MM raised a separate concern about its error model. An independent flat-ΛCDM fit using MM alone produced χ2/dof = 127.5/12 and p = 1.9 × 10−21. At z = 0.55 and 0.57, angular-scale discrepancies were about 22% despite individual uncertainties of about 2%.
Using MM values and covariance in a later compilation produced similarly poor statistics: χ2/dof = 129.2/15, p = 3 × 10−20, with residuals of −6.6σ at z = 0.55 and +4.2σ at z = 0.39.
The analysis therefore inflated MM uncertainties by approximately 3.3 to represent the observed scatter. N20 did not require such inflation; its fit gave χ2/dof = 11.8/10 with p = 0.3.
A calibration warning, not a new-physics claim
Redshift-binned diagnostics found no significant evolution in ε. The main test assumes flat-FLRW geometry; in a joint fit allowing curvature and calibration, Ωk remained consistent with flatness, and the flat and best-fitting curved cases differed by Δχ2 < 2.
The paper interprets the pattern as a relative calibration or standard-ruler inconsistency between the BAO sectors, not as new physics, a dark-energy effect or an expected spatial-curvature signal.
The practical caution is narrower: a transverse-only fit can look statistically healthy while hiding a constant calibration shift, and an error model that understates scatter can make discrepancies appear more precise than the data support.
Paper data and sources
Original title: A blind spot in transverse BAO calibration
Authors: Domenico Sapone
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text