Could the tension relief at low redshift mask pathologies that only emerge when fitting CMB or nucleosynthesis data?

This is a sharp question about a real cosmological concern. Your intuition points to a genuine tension in model-building: if a solution relaxes the Hubble tension by modifying the *late* universe (low redshift), could it hide inconsistencies that only show up when you demand agreement with the *early* universe (CMB recombination and Big Bang nucleosynthesis)?

The archive suggests this is exactly what happens with several proposed fixes.

The core problem

Early-universe measurements (particularly Planck CMB) constrain the sound horizon — the distance sound waves could travel before recombination — with exquisite precision. Late-universe measurements (supernovae distance ladder) infer the Hubble constant *today*. If you try to raise the Hubble constant without also raising the sound horizon, you create a mismatch. Conversely, if you modify early-universe physics to raise the sound horizon, you risk violating CMB constraints, which are already very tight.

What the preprints show

The late-time relief problem: Solutions that activate only at low redshift (z ≲ 1) can indeed sidestep some early-universe pathologies. For example, a horizon entanglement deficit mechanism [AG-2026.01-473] "raises the late time expansion rate by a few percent without affecting recombination or the sound horizon." This is elegant precisely because it avoids rewriting the early universe. However, the cost is that you must introduce new low-redshift physics whose origin is unclear.

Early modifications are tightly constrained: When you try to solve the tension *through* early-universe physics — modifying the primordial power spectrum, for instance — the CMB itself becomes your constraint. Work on scale-dependent primordial perturbations [AG-2025.04-168] found that "solutions are associated with a low total matter density $Ω_m$, which makes them inconsistent with baryon acoustic oscillations (BAO) and uncalibrated supernovae (SNIa) data." In other words, you cannot satisfy CMB *and* low-redshift data simultaneously with these modifications. The CMB becomes the arbiter that rules out would-be solutions.

A stress-test from pure geometry: Most directly, a model-independent reconstruction [AG-2026.04-1527] shows that to resolve the Hubble tension while satisfying CMB constraints, the expansion rate must jump by ~15% around matter-radiation equality. This is not a pathology that hides — it is a *feature* that any working early-universe solution must produce. This constrains future model-building sharply.

The pragmatic picture

The short answer is yes, but the opposite direction often matters more. CMB and nucleosynthesis data are so constraining that early-universe modifications to solve the tension tend to *fail* those tests, not hide pathologies in them. If a late-time modification avoids the early universe, it buys you freedom — but then you must explain *why* new physics activates at z ≲ 1, and whether it creates tension with baryon acoustic oscillations or other probes.

A recent survey [AG-2025.03-066] concluded that "the statistical preference for [ΛCDM extensions] remains moderate" and hinted that "improved systematic understanding in the CMB sector may be required" — suggesting the problem may not be pathological new physics at all, but rather unaccounted-for measurement systematics. That is perhaps the most sobering lesson: the tension might reflect neither a failure of the early universe nor a hidden late-time monster, but simply that our confidence in current precision datasets has been overstated.