Hiding cosmic answers to help astronomers test their methods

Astronomers measuring the universe’s expansion need protection against more than instrumental errors. Their expectations can also influence decisions about how to process data. A study led by G. Perez-Sanchez presents and tests a way to hide the cosmological answer in quasar observations while researchers refine their analysis.

The approach is called blinding: deliberately modifying information so analysts cannot easily tell whether their choices move a result toward an expected answer. This study evaluates an analysis safeguard, not a new measurement of the universe’s expansion. First submitted to arXiv as a preprint on 8 July 2026 at 19:18:54 UTC, the paper was revised on 5 November 2026. Its arXiv record also lists publication in the Journal of Cosmology and Astroparticle Physics, JCAP 11 (2026) 013.

Reading the hydrogen forest

The method targets the Lyman-alpha forest, a collection of absorption features in quasar spectra. Quasars provide bright background light; intervening neutral hydrogen absorbs particular wavelengths, leaving a record of gas along each sightline. Cosmic expansion stretches those wavelengths, allowing researchers to assign redshifts—and, with an assumed cosmological model, distances—to the absorbing material.

Comparing absorption across many sightlines reveals large-scale structure. One useful feature is the imprint of baryon acoustic oscillations, or BAO: a characteristic separation inherited from sound waves in the early universe. It serves as a standard ruler for measuring cosmic distances.

The proposed disguise builds on the Alcock–Paczynski effect. Converting angles and redshifts into distances with an unsuitable expansion model can stretch inferred structure differently along and across the viewing direction. The researchers deliberately introduce a controlled version of this geometrical distortion.

Changing coordinates, not the universe

First, the analysis estimates each quasar’s continuum, the spectrum expected without intervening absorption. It then extracts fluctuations in transmitted light relative to the estimated mean. Blinding happens after that extraction, because the quasar’s own emission and the intervening absorption originate at different redshifts and cannot simply receive one common spectral shift.

The algorithm constructs a mapping between distances in a reference cosmology and an alternative, hidden cosmology. It uses that mapping to change the wavelengths assigned to the absorption fluctuations and consistently shifts the catalogued quasar redshifts. Subsequent distance calculations and fitting use the unchanged baseline analysis, causing the recovered BAO ruler to appear at a deliberately displaced position.

The transformed absorption samples no longer have uniform spacing. They are therefore resampled onto a regular grid, together with their statistical weights. This prevents the irregular spacing itself from revealing the transformation, although resampling slightly smooths features and changes weights.

Applying the disguise this early also means that later stages—including correlation measurements and model fitting—can operate on blinded inputs, rather than hiding only the final parameter estimates.

Tests from simulations to observations

The team implemented the method within PICCA, software used for Lyman-alpha forest analysis. Validation began with simulated catalogs containing more than a million quasars and very little observational noise. Tests first included hydrogen absorption alone, then added other absorption lines. A more realistic simulation contained approximately 420,000 quasars with observational noise.

Finally, the researchers applied the procedure to the Dark Energy Spectroscopic Instrument’s first data release, using spectra from more than 420,000 quasars at redshifts above 2.1. They examined both correlations within the absorption field and cross-correlations between absorption and quasar positions.

Tests explored relative changes in the assumed matter-density parameter from a 5% decrease to a 10% increase. These are chosen disguises, not measured changes in cosmic matter content. The simulated BAO shifts agreed with the predicted shifts within the reported one-standard-deviation uncertainties. The unblinded observational analysis also closely reproduced the published DESI baseline, providing a check on the implementation.

Probability distributions for the fitted BAO parameters generally retained similar shapes, though the paper reports small changes in widths and orientation. The procedure added approximately 5% to total processing time in the tested workflow.

Where the disguise can leak

A key limitation comes from overlapping absorption signals. The mapping treats all absorption as Lyman-alpha, but real spectra also contain Lyman-beta absorption and lines from heavier elements, called metals in astronomy. These contributions cannot readily be separated before applying the transformation.

Known metal-line features can consequently reveal clues in one-dimensional correlations, which compare absorption along individual sightlines. Their displaced positions may indicate the direction of the blinding shift, although they do not directly reveal the full transformation. The authors suggest additional blinding at a later analysis stage to reduce this leakage.

The validation also focused on BAO positions while holding other model parameters fixed. It therefore does not establish performance for every analysis of the full clustering pattern. If statistical uncertainties are estimated from separate simulations, those simulations must also remain sufficiently close to both the blinded and unblinded cosmologies for the uncertainty estimates to remain representative.

Larger modern simulation suites and later DESI releases remain future tests. The demonstrated result is narrower but useful: a coordinated disguise can shift the intended cosmological ruler while largely preserving the tested analysis’s statistical behavior. Using the same transformation for other tracers could also support jointly blinded analyses, an extension demonstrated here for the forest–quasar cross-correlation rather than for every possible combination of observations.

Sources: the arXiv paper and submission history, and the journal publication listed in that record.