Where It All Began / Cosmology tests
THE ORIGIN QUESTION

Where It All Began

Origin Question ranks the leading explanations for how the universe began. We weigh evidence from any source readers can verify and update the leaderboard as new findings come in.

Five tests of the universe’s origin

View the leaderboard →

Could recent evidence change the standings? We ran five checks with published measurements and model predictions. The strongest candidate for a future score change is evolving dark energy. These runs give us a reason to keep watching it, while all five leaderboard percentages still match their starting values.

The calculations use published summary statistics and simplified models. Measurements, correlations and instrument effects enter through the cited papers. The three Monte Carlo checks use 1,000,000 draws per simulated distribution, starting from seed 20261116. Two other checks sweep a fixed grid.

Download the runnable calculations and inputs

1. Does dark energy change over time?

The DESI fit improves when dark energy can evolve. How much does that improvement depend on which supernova catalog is used?

How we ran it We simulated the null distribution for a two-parameter extension of the cosmological constant and compared it to the fit improvements from DESI Data Release 2, Table VI. The χ² approximation gives a tail area exp(−Δχ²/2), where Δχ² is the positive fit improvement shown here. The samples share galaxy and cosmic microwave background (CMB) data.

How often the simulated constant model produces an improvement this large
Data used in the published fitFit improvement Δχ²Simulated countApproximate null tail
DESI + CMB (no supernovae)12.51,8930.193%
+ Pantheon+10.74,6980.475%
+ Union317.41770.0167%
+ DESY521.0350.00275%

Leaderboard decision Keep 17% for now. Across these DESI fits, the preference survives changing the supernova catalog, but its strength varies. The 2026 Supernovae Unite preprint, combining and recalibrating overlapping catalogs, reports 2.5σ using the posterior maximum, 3.1σ using maximum likelihood, and weak Bayesian evidence. A clear independent measurement could justify moving this from 17% to 29% under the current rubric.

2. Which inflation models fit the sky?

A plateau model and a quadratic model can give similar scalar fluctuations but predict very different amounts of primordial gravitational waves.

How we ran it We evaluated 201 equally spaced durations from 50 to 60 e-folds for each model. At leading slow-roll order, both have scalar tilt nₛ ≈ 1 − 2/N. Their tensor ratios are r ≈ 12/N² (Starobinsky) and r ≈ 8/N (quadratic). We screened against Planck’s nₛ = 0.9649 ± 0.0042 at about 95% and the BICEP/Keck bound r < 0.036. The two separate cuts are a screening rule rather than a joint likelihood.

Inflation models screened against both observational cuts
ModelPoints passingPredicted tensor ratio r
Starobinsky201 / 2010.0033 to 0.0048
Quadratic0 / 2010.1333 to 0.1600

Leaderboard decision Keep 33%. The quadratic model fails this screen throughout the tested range, while the Starobinsky model passes. The distinction narrows which forms of inflation are viable.

3. Would Planck have seen these bounce signals?

Three published bounce templates predict a measurable three-way pattern in the ancient light of the universe. We checked whether the predicted amplitudes should have been recoverable.

How we ran it Using Tables I and III of van Tent, Delgado and Durrer, we simulated Gaussian estimates of each template amplitude with the published Planck uncertainty. We injected the amplitude needed to make the large-scale anomaly a 20% occurrence, then counted absolute estimates three standard deviations from zero. We also computed the observation-to-prediction gap for weaker 5% variants.

Simulated recovery and the actual Planck comparison
Bounce templatePlanck estimate ± 1σSimulated 3σ recoveryGap for 20% variantGap for 5% variant
q = −0.5240 ± 470990,660 / 1,000,0005.9σ3.2σ
q = −0.7 (loop quantum cosmology)160 ± 260999,639 / 1,000,0007.0σ3.9σ
q = −1.2419 ± 341,000,000 / 1,000,00014.7σ7.8σ

Leaderboard decision Keep 0%. The strongest injected signals would usually have shown up, and Planck saw amplitudes near zero. Our gaps are measured from the fitted Planck amplitudes, while the paper’s quoted 5.4σ, 6.4σ and 14σ are approximately the nominal signal relative to zero. These results weigh against the three templates and their chosen amplitudes. Other bounce mechanisms can predict different signals.

4. How much room is left for a cyclic model?

In one family of cyclic models, a contracting phase can leave an imprint on how early density fluctuations are distributed.

How we ran it We used the kinetic-conversion approximation in Planck’s non-Gaussianity paper, fₙₗ ≈ 1.5 κ₃√ε ± 5. We fixed ε = 100 as an illustrative benchmark and swept κ₃ from −5 to 5 in steps of 0.01 for both conversion branches. We kept predictions within 1.96 standard deviations of Planck’s temperature-and-polarization measurement, −0.9 ± 5.1.

Compatible parameter values in the scanned grid
Conversion offsetAllowed κ₃ valuesGap at illustrative κ₃ = 3
+5−1.05 to 0.2710.0σ
−5−0.39 to 0.938.0σ

Leaderboard decision Keep 0%. The test rejects large signals in this benchmark and leaves small signals possible. The result constrains the chosen conversion model and depends on ε. It supplies no positive detection of a previous cycle.

5. Did the hot Big Bang make the right amount of deuterium?

The microwave background and ancient gas clouds offer different ways to check how much deuterium the early universe produced.

How we ran it We used the published LUNA-based nucleosynthesis prediction of 2.52 ± 0.03 (baryon density) ± 0.06 (nuclear rates), and Cooke et al.’s observed 2.527 ± 0.030, all in units of 10⁻⁵ for the deuterium-to-hydrogen ratio. We drew independent Gaussian errors and calculated observed minus predicted in each trial.

Deuterium-to-hydrogen ratio in units of 10⁻⁵
PredictedObservedGap in combined standard deviationsCentral 95% of simulated differences
2.522.5270.10−0.137 to 0.151

Leaderboard decision Keep 50%. The observed abundance differs from the prediction by about one tenth of a combined standard deviation. The hot Big Bang keeps its current evidence grade.