BigSnap CLASS Tests — Series 2: The Universe is Clumpier in the Past Than Today — and We Can Measure Why

The standard model of cosmology has a problem. When astronomers measure how clumped together matter is in the universe today, they get a lower number than what the Big Bang predicts. Two different measurements of the same universe disagree. Nobody has a clean explanation for why.

This paper tests a specific idea: what if dark matter has a tiny amount of pressure that only switches on in the last few billion years? Early in the universe — no pressure, everything clumps normally. But recently, as dark energy started dominating over dark matter, the pressure slowly built up and started resisting gravitational collapse. Less clumping. Lower number. Gap closed.

We ran this idea through CLASS — the same computer code used by the Planck satellite collaboration and major cosmological surveys to test theories against data. We asked: does this pressure model produce the right amount of suppression? Does it break anything else? And critically — can you tell this apart from the most obvious alternative explanation, which is that neutrinos (tiny particles with mass) are doing the suppression instead?

The key result: the pressure model and the neutrino model both produce the same overall suppression today. But they do it differently over time. Neutrinos suppress structure at every point in cosmic history equally — from 13 billion years ago to now, always the same. The pressure model does almost nothing early on and then ramps up only in the last few billion years. If you look at the universe in distance slices, the pressure model shows strong suppression in the nearby slices and almost no suppression in the distant slices. Neutrinos show the same suppression in every slice. That difference is the tomographic tilt — a specific, measurable prediction about how the clumpiness of the universe changes with distance.

We threw realistic measurement errors at the tilt signal and asked how much survives. About 22% survives even in the worst case. More importantly, when combined with a completely different kind of observation — how fast galaxies are moving apart, measured from their spectra — the two signals point in the same direction while their measurement errors point in different directions. That means the combined signal is very hard to fake.

The model makes a specific, falsifiable prediction: the S₈ discrepancy should be strongest when you look at nearby galaxies and should disappear when you look at distant ones. This prediction is stated here before any comparison to real survey data is performed.

What the results actually show — without bias:

The model passes every internal consistency check. It does not break the CMB, it does not conflict with current growth rate measurements, and it does not shift the BAO scale. The suppression it produces is in the right ballpark to address the observed discrepancy.

However, the model does not yet have a full likelihood analysis behind it. We have not run it against the combined weight of Planck, KiDS, DESI, and BOSS simultaneously. The tomographic tilt signal — while real in the model — is partially absorbed by realistic measurement errors, and only about a fifth of it survives worst-case contamination. The joint probe test improves this, but it too is an optimistic bound pending a proper covariance-weighted analysis.

The honest summary is this: the model is consistent with everything we have tested, it makes a distinctive prediction that differs from neutrino mass, and that prediction is testable with current and near-future surveys. It has not been proven right. It has not been ruled out. It has earned the right to be tested.

All numerical results, CLASS implementation details, parameter scans, validation tables, stress test outputs, and figures are included in the downloadable documents with this deposit.

Companion papers: BigSnap Framework (Snider 2026), CLASS Series 1 — Implementation and First Results (Snider 2026).