Reanalysis Challenges DESI’s Claimed Signal for Dynamical Dark Energy

A new arXiv preprint argues that one of cosmology’s most closely watched hints of dynamical dark energy in DESI data is not robust. The authors say the apparent signal in the Dark Energy Spectroscopic Instrument’s latest baryon acoustic oscillation, or BAO, results is concentrated almost entirely in a single redshift bin and can disappear under reasonable alternative statistical checks.

The paper, “A Sequentially-Valid Reanalysis of DESI’s Dynamical Dark Energy Signal,” was posted July 31 as arXiv:2607.28918v1 and is marked as a preprint “Prepared for submission to JCAP.” It is by Jinyoung Kim, David F. Mota and Andrius Tamosiunas, whose listed affiliations include Stanford University and the University of Oslo.

The reanalysis matters because DESI’s 2025 DR2 cosmology paper helped fuel interest in the idea that dark energy — the unknown component driving cosmic expansion — might change over time rather than behave like a simple cosmological constant, as in the standard ΛCDM model. In that paper, the DESI collaboration said of one key combination, “This solution is preferred over ΛCDM at 3.1σ for the combination of DESI BAO and CMB data.”

DESI is a major sky survey, and its DR2 BAO analysis used measurements from more than 14 million galaxies and quasars to trace the universe’s expansion history. A credible departure from ΛCDM would be a major result, which is why follow-up checks of the statistics have drawn attention.

Kim, Mota and Tamosiunas revisit the sequence from DESI’s first BAO release to its second using public DR2 BAO data products. Their main point is that DESI data arrive in stages, and researchers are likely to revisit the same dark-energy question repeatedly as DR1, DR2, DR3 and later releases appear. Using ordinary one-shot sigma values at each step can make evidence look stronger than it is because the data are being checked multiple times.

To address that, the authors use what they call an e-process, which they describe as “an e-process … remains valid under repeated looks at accumulating data.” In plain terms, it is a way to track evidence over time without overstating significance just because analysts keep checking updated data releases.

Under the paper’s default test setup, aligned with the direction favored by DR2, the authors report a strong-looking result: “the running evidence reaches M_DR2 = 33.97, crossing the threshold of 20 that corresponds to an illustrative 5% false-detection rate.” But the same analysis says that evidence is “concentrated almost entirely in the LRG2 redshift bin,” referring to one of the seven BAO bins.

That concentration turns out to be crucial. When the LRG2 bin is removed, the running e-value drops to M = 0.49, which the authors say mildly favors ΛCDM instead of dynamical dark energy. They also report that if analysts allow for the possibility that any one of the seven BAO bins could have produced the excess, the signal does not survive a look-elsewhere correction.

The paper further argues that the result is sensitive to how the test is specified. Some alternative setups do not reject ΛCDM. The authors also find that a physically motivated thawing-quintessence model can reject ΛCDM more strongly than more agnostic alternatives, but they add that a signal concentrated in one bin is difficult to reconcile with a smoothly evolving dark-energy equation of state.

Their bottom line is not that DESI’s earlier result has been disproven, but that the current evidence in the DR1-to-DR2 BAO sequence looks fragile, single-bin and specification-dependent rather than robust. That broadly fits with several earlier follow-up papers that have suggested the apparent preference for evolving dark energy may depend on priors, dataset choices or particular galaxy bins.

If a genuine departure from ΛCDM is eventually established, it would reshape one of the central assumptions of modern cosmology. For now, Kim and colleagues argue that future releases from DESI, Euclid, NASA’s Roman Space Telescope and the Vera C. Rubin Observatory’s LSST should report anytime-valid e-process values alongside conventional sigma significances, so readers can better judge how evidence holds up as the data keep accumulating.