Is the Hubble tension a real discrepancy in the expansion rate of the universe, or a measurement systematic?

There are two honest ways to weigh the expansion of the universe, and they begin at opposite ends of time. The first starts 380,000 years after the Big Bang, when the cosmos cooled enough for light to travel freely. That first light — the cosmic microwave background — carries a pattern of ripples whose sizes are set by well-understood physics. The Planck satellite mapped the pattern; feed it into the standard model of cosmology and run the clock forward, and today's expansion rate comes out at 67.4, give or take 0.5, in the units astronomers use: kilometres per second of recession speed, per megaparsec of distance. ◆4 families It is a beautiful measurement with one honest asterisk: it is an inference through a model, not a direct reading — change the model, and the number moves. ◆3 families

The second way ignores the early universe entirely and climbs a ladder in the dark, rung by rung. Geometry fixes the first rung: parallax and other direct methods give distances to nearby pulsing stars called Cepheids. Cepheids calibrate the second rung, because their pulse rate reveals their true brightness. And Cepheids, in turn, calibrate exploding white dwarfs — Type Ia supernovae — bright enough to be seen across the cosmic gulf. Each rung inherits the errors of the one below it. ◆3 families The SH0ES program has spent two decades hardening this ladder, and it reads 73.0, give or take 1.0. ◆4 families

Neither camp stands alone. Galaxy-survey measurements calibrated by the physics of the early universe — but using no Planck data at all — come out low, near 68.5. ◆3 families The early universe, asked politely and asked twice, keeps saying slower. The local universe keeps saying faster.

Set the two headline numbers side by side and the gap is 5.6 — about nine times the CMB measurement's stated uncertainty. Folding both error bars together, the disagreement stands at roughly five sigma: if the uncertainties are honest and complete, chance alone produces a gap this large about once in a million tries. ◆5 families The council's re-derivation verifier recomputed that significance from the published numbers rather than trusting anyone's abstract; it checks out.

Five sigma is not itself the finding — the finding is that the gap will not move. Swap the supernova sample, swap the geometric anchors, swap the analysis pipeline: the ladder's answer shifts by fractions of a unit while the gap needs 5.6. ◆3 families And the ladder is not the only local voice. Water masers orbiting distant black holes, and the delays between multiple images of gravitationally lensed quasars, mostly land high as well — though the council notes, honestly, that the lensing numbers loosen considerably once you stop assuming a particular mass profile for the lens, and some of these methods can live with either camp. ◆3 families

What has not appeared, in a decade of audits, is any single identified error big enough to close the gap on its own. ◆3 families That is what the council converged on — not that anyone knows the answer, but that the question is real. The discrepancy is not an artifact of one team, one instrument, or one bad night at the telescope. Something, somewhere, is genuinely unaccounted for.

If one of the measurements is subtly wrong, most suspicion has always pointed at the ladder — not because the ladder team is careless, but because a ladder has more places to hide an error. The longest-favoured suspect was crowding: Cepheids sit in dense star fields, and if faint neighbours contaminated their photometry, every rung above would inherit the bias. JWST settled it. Its sharper infrared eye re-photographed the same Cepheids and reproduced the Hubble-telescope measurements almost exactly. ◆4 families The specific claim that crowding explains the tension now stands contradicted on the board, and the family that carried it withdrew with the citation in hand. ◇1 family

The live dispute is about the calibrator stars themselves. Swap the Cepheids for a different standard candle — the tip of the red-giant branch, a feature of old stars that lights up at a predictable brightness — and the answer drifts down toward 69.8, parked awkwardly between the camps. ◑contested This is where the council genuinely split. One program, averaging three independent stellar calibrators on JWST data, lands close to 70; the ladder team's analysis of overlapping data from the same telescope finds no departure from 73. ◑contested Same instrument, same sky, two defensible answers — the disagreement now lives in choices about which galaxies to include and how to combine them, and the record holds both readings at full strength.

What remains is bookkeeping that refuses to finish. How supernova light is corrected for dust in its host galaxy could plausibly move the answer by a unit or so, and no end-to-end reanalysis has settled the size of the effect. ◇unverified But every named systematic, taken alone, has been bounded well below the size of the gap — so the systematics case now requires several independent errors, each individually small, all leaning the same way. ◆supported Not impossible. Not comfortable either.

Suppose instead that both measurements are right. Then the fault is in the model that connects them — and there is surprisingly little room to hide. The early-universe numbers rest on a single yardstick: the distance sound waves could travel in the primordial plasma before the first light was released. Late-time fixes — tinkering with dark energy after the fact — cannot rescue both the CMB and the galaxy-survey data at once. The arithmetic pushes any real solution into the first 380,000 years: something must shrink that yardstick. ◆3 families

The best-developed candidate is early dark energy — a brief, self-erasing burst of expansion pressure just before the first light. It can lift the early-universe prediction to around 70 or 72, but it buys that relief at a price: the fit to how galaxies later clustered gets worse, and the new ingredient must switch on and off at a suspiciously convenient moment. ◆3 families On the board, the harder sentence is the general one: no proposed extension of the standard model currently resolves the tension without degrading the fit somewhere else. Most families hold that claim; one maintains that a handful of early-universe variants remain statistically viable, and its dissent is open. ◑contested

The way out is more measurement. Gravitational waves from colliding neutron stars arrive with their distance written into the waveform — no ladder, no yardstick — and with a hundred well-placed events they should read the expansion rate to a percent. ◇2 families Until then, the council's crux claim stays split down the middle: asked whether this ends in an overlooked systematic or in new physics, the families genuinely disagree, and the record says so rather than pretending otherwise. ◑contested The tension is real. Its meaning is not yet earned.