Measurement & Units

Part Ithe lineage

Follow the digits down

Take c first. In 1983 the CGPM fixed the speed of light at 299 792 458 m/s. That was the best measured value at the time—measured, of course, in metres. The metre of 1983 was the krypton metre of 1960, built to match the platinum-iridium bar of 1889, made as faithfully as possible from the Mètre des Archives of 1799, cut from the survey that Delambre and Méchain dragged across France between 1792 and 1798—the survey with a concealed discrepancy in it.

The numerical value 299 792 458 is therefore not a unit-free fingerprint of light. It preserves the best measured value of c in the inherited metre and second, whose lineages reach through the krypton bar and astronomical time to a French survey and observations of the Moon. The physics is light; the digits also record Méchain and the clocks that followed him.

The kilogram’s ancestry is wetter. The second’s is stranger: its last digits encode the motion of the Moon, photographed between 1954 and 1958. Pick a unit and drill.

The second’s lineage contains the appendix’s most important move. In 1958, Markowitz, Hall, Essen, and Parry reported the caesium frequency against Ephemeris Time as 9 192 631 770 ± 20 cycles per ephemeris second. It was a frequency measurement with an uncertainty.

In 1967, the 13th CGPM took the measured number, discarded the ±20, and declared the remainder exact. That was the 2019 reform fifty-two years early. The uncertainty did not evaporate; it was absorbed into the definition, just as the uncertainty in h was absorbed in 2019 and the uncertainty in c in 1983. Every time we make a constant exact, we repeat the same rite: freeze the final measured value and move its uncertainty elsewhere.

A defining constant freezes a measured numerical value so the unit remains continuous. The physics—and every dimensionless comparison—remains open to test.

Part IIthe unit laboratories worked around

The ampere’s twenty-nine-year secret

Here is the pre-2019 definition of the ampere:

The ampere is that constant current which, if maintained in two straight parallel conductors of infinite length, of negligible circular cross-section, and placed 1 metre apart in vacuum, would produce between these conductors a force equal to 2 × 10⁻⁷ newton per metre of length.

Infinite length. Negligible cross-section. Vacuum. No apparatus could instantiate those idealizations literally. Finite current balances could approximate them with calculated corrections, but the definition was extraordinarily difficult to realize accurately. It remained in international measurement law for seventy-one years, from 1948 until 2019.

Electrical metrologists therefore seceded. On 1 January 1990, following a CIPM recommendation, they adopted conventional values—deliberately and explicitly not the SI—for the two quantum constants they actually used:

• $K_{J-90}=483597.9GHz/V$, the Josephson constant;
• $R_{K-90}=25812.807\Omega$, the von Klitzing constant.

Both were assigned zero uncertainty, creating a parallel system of the conventional volt V₉₀, conventional ohm Ω₉₀, and conventional ampere A₉₀. For twenty-nine years, every precision voltmeter was calibrated in a currency close to SI legal tender but not technically legal tender. Precision electrical measurements had lost formal traceability to the fundamental physical constants.

The 2019 reform ended the secession. The 26th CGPM abrogated the conventional values, defined the ampere through e, and legalized what electrical laboratories had been doing all along.

The reform was designed so that units would not visibly change. That promise held everywhere except the domain that had gone outside the old system. On 20 May 2019, voltage-related quantities became 1.067 × 10⁻⁷ larger and resistance-related quantities 1.779 × 10⁻⁸ larger. The step was invisible in ordinary life and seismic to anyone calibrating voltage references. The reform’s one visible disruption fell precisely where the old definition had been least usable.

Part IIIwhy any of this works

The metrologist’s best friend is an integer

In the Kibble-balance route, the kilogram is realized through electrical standards based on the Josephson effect and the quantum Hall effect. The independent silicon-sphere route uses different experimental physics. Why can a superconducting junction or a semiconductor sample—imperfect objects made from different materials in different laboratories—nevertheless reproduce h/2e and h/e² to ten significant figures?

Their outputs are counted, not measured.

Brian Josephson predicted the first effect in 1962, as a 22-year-old graduate student, and received the 1973 Nobel Prize. Shine microwaves of frequency f at a superconducting junction and its voltage locks onto steps,

$U=nf\frac{h}{2e}$,

where n is an integer. Within a valid Josephson operating step, its nominal voltage depends on the microwave frequency and two defining constants rather than on material or geometry. A real realization still carries uncertainty from frequency, wiring, step identification, and whether the junction remains in the required operating regime.

Klaus von Klitzing found the same pattern in 1980 and received the 1985 Nobel Prize. Put a two-dimensional electron gas in a strong magnetic field and the Hall resistance climbs a staircase,

$R=\frac{R_K}{i}=\frac{h}{i{e}^2}$,

with i an integer. Silicon, gallium arsenide, and graphene give the same plateaux and the same heights to parts in 10¹⁰.

The quantization is topological. Within the appropriate low-temperature, high-field, non-dissipative regime, the integer invariant cannot drift by 0.1%; it can only change by leaving that regime or moving to another plateau. Real standards therefore verify plateau selection, contact resistance, current limits, and the absence of breakdown.

The useful twist is that a cleaner sample does not give better plateaux. Without disorder, there would be no broad plateaux at all. Disorder localizes electronic states between Landau levels and gives each plateau a finite width—a flat shelf on which a measurement can sit. A perfectly pure crystal would supply exact quantization only at isolated field values: correct and useless. The imperfection is load-bearing.

The result supplies the hidden spine of the 2019 reform: build standards whose ideal outputs are discrete and protected. An artefact can shed a fingerprint’s worth of platinum; a lamp can run slightly warm. A quantum standard can still be misidentified, miscounted, or driven outside its operating regime, but those failure modes can be tested rather than silently redefining the unit.

The SI has run relentlessly toward counting. The 1960 metre counted wavelengths. The 2019 mole counts entities. The silicon-sphere kilogram counts atoms. The Josephson volt identifies steps; the quantum Hall ohm identifies plateaux; the second counts oscillations. Every escape from the artefact has leaned harder on discrete structure plus checks that the apparatus is realizing the intended state.

This is Day 7 returning from an unexpected direction. Shannon taught that information is discrete and physical. Metrology adds the corollary: discreteness can make a standard robust, because a validated integer step is less vulnerable to small material drift than an artefact’s continuously changing surface.

Part IVthe calibrated arithmetic of error

Precise, true, or accurate? They are three different things

The familiar archery-target picture is often taught backwards. The distinction matters:

• Precision is the tightness of the grouping: how much repeated answers scatter.
• Trueness asks whether the centre of the grouping sits on the bullseye: it concerns bias.
• Accuracy requires both.

The asymmetry is uncomfortable: precision is visible from inside your laboratory; trueness is not. Repeat a measurement a thousand times and you learn how well the apparatus agrees with itself. Those repetitions alone say nothing about whether it agrees with the world. The most dangerous quadrant is therefore not a messy cloud but a tight cluster confidently displaced from the bullseye. It looks publishable.

The main descent’s α dipole illustrates the danger without supplying a clean textbook verdict. Its published error budget omitted long-range wavelength distortions comparable to the claimed signal. The 2015 audit did not prove that every possible dipole is false; it removed the basis for treating this one as established. More observations with the same uncorrected instruments could have narrowed an interval around an artefact, while an external calibration audit exposed the missing systematic.

The judgment inside every uncertainty budget

The GUM—the Guide to the Expression of Uncertainty in Measurement—governs how calibration certificates are written. It did not abolish error. It distinguishes the unknowable error of a particular result from the uncertainty used to characterize the values reasonably attributable to the measurand.

“Error” is the difference between a measured value and a reference quantity value, but the actual error normally remains unknown. Uncertainty instead characterizes the dispersion of values that could reasonably be attributed to the measurand. A report may give a standard uncertainty or an expanded interval with stated coverage assumptions. Neither is direct access to truth, and not every reported uncertainty is a frequentist confidence interval.

It then divides evaluation into two kinds:

• Type A uncertainty is evaluated statistically from repeated observations. It describes countable variation and precision.
• Type B uncertainty is evaluated by other information: a calibration certificate, manufacturer’s data sheet, handbook value, physical model, prior experience, and ultimately the considered judgment of an expert.

Type B is a route of evaluation, not a declaration that every input is a subjective Bayesian prior. Its probability distributions can come from calibration data, physical models, published specifications, or expert judgment, depending on the case. Once expressed as standard uncertainties, Type A and Type B contributions are propagated through the same uncertainty budget. The important epistemic point is narrower and more durable: even a reported 5.4-sigma discrepancy can fail when the uncertainty model omits an instrumental bias, and a reported “±” can depend on judgments about models, calibrations, and possible systematics as well as on repeatability statistics. That is why the Berkeley–Paris α tension remains unresolved and why an unidentified systematic can hide inside a beautiful error bar for years.

Part Vthe other four

Listening to a gas, counting a dozen, and the human eye hiding in the SI

The kelvin: determine Boltzmann’s constant by listening

The leading route to k was acoustic gas thermometry. Fill a precisely machined resonant cavity with argon, ring it, and measure the speed of sound u. Because an ideal gas obeys $u^2=\gamma kT/m$, the pitch of the gas reveals Boltzmann’s constant. Independent checks came from dielectric-constant gas thermometry, which reached about 1.9 × 10⁻⁶, and Johnson-noise thermometry, which extracts k from the thermal hiss across a resistor. Three unrelated physical routes converged on one number.

The new kelvin’s real payoff appears at the extremes. The old definition anchored thermodynamic temperature to the triple point of water, 273.16 K; primary thermometry and practical scales then disseminated temperature through multiple methods and fixed points, with the traceability chain becoming less direct far from the anchor. The new kelvin permits primary realization at the temperature of interest. The reform changed nothing in a kitchen and opened better routes at the extremes.

The mole: a dozen, writ large—and chemists objected

Since 2019, a mole is exactly 6.022 140 76 × 10²³ entities. It is a count, decoupled from the kilogram. Some chemists raised a sharp objection: if the mole is only a number, is “amount of substance” really a base quantity, or a dimensionless count wearing a costume? A dozen is not a base unit. Why is a mole? The dispute reaches beyond pedantry to whether the seven-base-unit SI reflects nature or the history of its committees.

The candela: an alien could never derive it

The candela fixes $K_{\mathrm{cd}}=683lm/W$ for monochromatic radiation at 540 × 10¹² Hz, green light near 555 nm. The frequency is the peak of V(λ), the photopic luminosity function: the standard human eye-response curve adopted by the CIE in 1924 from a modest set of observers. Photometry weights every other wavelength by V(λ).

Among the speed of light, quantum of action, and elementary charge, the seventh defining slot therefore canonizes the average sensitivity of a human retina. Even 683 was reverse-engineered for continuity with candle-based standards. An alien civilization given our seven defining numbers could reproduce all seven SI units, but it could derive none of their historical scale choices from physics alone. The candela is distinctive because its definition adds a human visual-response function, not because the other six numerical scales are inevitable.

So why seven? The careful answer might be zero

History supplied the count. Maxwell proposed length, mass, and time; Giorgi argued in 1901 that electromagnetism needed a fourth unit; the mole and candela accreted for chemistry and lighting. The base/derived split is a bookkeeping convention rather than a discovery.

Planck units push in the other direction by setting c, ħ, G, and often k to one: a length near 1.62 × 10⁻³⁵ m, time near 5.39 × 10⁻⁴⁴ s, and temperature near 1.42 × 10³² K. Yet the Planck mass is about 22 micrograms, roughly a flea’s egg—strangely human-scale beside an unimaginably small length and enormous temperature. That juxtaposition is not itself the hierarchy problem; the physics problem concerns the enormous gap between the electroweak and Planck scales, which Day 45 will revisit. Duff’s question remains: are Planck units nature’s units or simply a cleaner arbitrary grid?

Part VItime’s messy edges

The second collides with the shape of Earth

Earth is a poor clock. Lunar tides slow rotation by roughly 1.7 milliseconds per century. Core and mantle exchange angular momentum; the crust still rebounds from ice ages that ended about ten thousand years ago; the atmosphere moves with El Niño; earthquakes alter the moment of inertia. The wobble is irregular.

Timekeepers therefore maintain three times. TAI counts caesium ticks without consulting the sky. UT1 measures Earth’s actual rotation angle, which determines astronomical noon. UTC runs on atomic seconds but inserts a leap second whenever needed to stay within 0.9 seconds of UT1. Twenty-seven leap seconds have been inserted since 1972; the latest was on 31 December 2016.

Leap seconds cannot be scheduled far ahead—notice arrives only about six months in advance because Earth does not rotate predictably—and they have disrupted computer systems. In November 2022 the CGPM resolved to increase the permitted difference between UT1 and UTC by or before 2035, with the intended consequence that leap seconds cease for an extended period. “Abolish leap seconds” is useful shorthand, not the resolution’s literal wording.

Then Earth sped up. The years 2020–2024 included exceptionally short days, raising an unprecedented possibility: a negative leap second, a minute containing 59 seconds. Most leap-second software assumes the adjustment will be positive; the opposite direction has never been deployed.

Part VIIspending entanglement

The floor beneath the floor—and why squeezing is not free

Quantum mechanics now changes from the thing being measured into the thing being spent.

Interrogate N independent atoms and each readout collapses to “excited” or “not excited.” Independent quantum outcomes make frequency uncertainty fall only as $1/\sqrt{N}$. This quantum projection noise establishes the standard quantum limit. It is a limit on uncorrelated particles, not on lasers, vacuum systems, or budgets.

The escape is to correlate the atoms. A spin-squeezed state makes one collective component quieter by making another noisier. In principle, the limiting scaling improves from $1/\sqrt{N}$ toward $1/N$, the Heisenberg limit. Entanglement becomes a resource.

It works. Eckner and colleagues used Rydberg interactions to squeeze a programmable strontium clock array, obtaining about four decibels of metrological gain, stability 1.087(1) × 10⁻¹⁵ at 1 s, and performance 1.94(1) dB below the SQL (Nature 621: 734–739, 30 August 2023). In 2025, two spin-squeezed ensembles of about 30 000 strontium atoms in a two-dimensional lattice, read through cavity-QED quantum non-demolition measurements, reached fractional frequency precision of 1.1 × 10⁻¹⁸ with a 2.0(2) dB gain over the SQL (Physical Review Letters 135: 193202, 7 November 2025), the most precise entanglement-enhanced clock reported in the source material.

Open questions

Still unsettled

• Is “amount of substance” a genuine base quantity, or is the mole a dozen with a Nobel Prize?
• Should the candela remain a base unit when its weighting rests on a 1924 average retina?
• Can the global geoid ever be known to a centimetre? If not, maps rather than clocks may limit the redefined second.
• Should the new second rest on one optical transition—simple but brittle—or a weighted ensemble—robust, inelegant, and more future-proof? A BIPM draft resolution dated 30 June 2026 records that no consensus has been reached and asks the community to submit a redefinition proposal to the 29th CGPM in 2030.
• Does fixing a constant stop us from discovering variation? No: only dimensionless ratios carry unit-independent claims, and those remain measurable. The important task is understanding why.
• If exact numerical values are fossils of our history, an alien civilization would use different digits for the same physics. Which parts of science describe the world, and which record that we were the measurers?

Closing

The ghost in the machine

The 2019 SI is a cathedral of invariants. Nothing in it can rust, be stolen, gather fingerprints, or be dropped by a nervous technician. It is among the most durable things bureaucracy has built.

Yet at the bottom of its core sample lie:

• a mis-surveyed meridian through Revolutionary France, including a discrepancy that Méchain concealed and that consumed him;
• a cubic decimetre of water, weighed in 1799 and off by 28 parts per million;
• four years of photographs of the Moon, taken between 1954 and 1958 and checked against a table of the Sun computed before the electron was known.

We did not escape history. We encoded it. The constants are exact now, but they are exactly our numbers: values chosen to preserve continuity with prior standards and measurements within the transition uncertainties. That is not a defect. It is the system.

A unit is therefore a treaty across space and time. The refusal to round c promises everyone who measured before us that the meaning of their numbers will not silently change. Méchain hid a discrepancy and it broke him. Two centuries later, metrology understood it and, rather than changing the size of the metre, preserved the continuity of everything built above it.

Part Ithe clock species explosion

Everyone is building a different clock now

For two decades, optical clocks were largely a two-horse race: neutral atoms in a lattice, especially strontium or ytterbium, versus one trapped ion, especially aluminium or ytterbium. Since 2020 the field has diversified because the aim is no longer only timekeeping. Different clocks respond to different possible new physics, so a varying constant or dark-matter field may reveal itself when a menagerie of clocks disagrees.

Clocks made of stripped atoms

Remove thirteen of argon’s eighteen electrons and Ar¹³⁺ remains: a highly charged ion whose electrons sit deep in the nuclear field. It is comparatively insensitive to stray fields that trouble other clocks, while some transitions in highly charged ions are exceptionally sensitive to α.

The obstacle was cooling and reading an ion with no accessible cooling transition. In 2022, Steven King, Lukas Spieß, Piet Schmidt, and colleagues at PTB co-trapped Ar¹³⁺ with a beryllium ion, sympathetically cooled it, and read it through quantum logic. The first HCI optical clock achieved systematic uncertainty of 2.2 × 10⁻¹⁷ on its first demonstration (King et al., Nature 611: 43–47, 2 November 2022). Comparing ³⁶Ar¹³⁺ and ⁴⁰Ar¹³⁺ also resolved the QED contribution to nuclear recoil in a many-electron system for the first time.

HCI clock built α-search payoff The clock demonstration is established. Its present uncertainty is roughly two dozen times worse than the best Al⁺ clock result, and no HCI clock has yet set a competitive new-physics bound. The attraction is a higher possible ceiling and stronger sensitivity, not a result already delivered.

Clocks made of molecules

Atomic clocks are highly sensitive to α. The proton-to-electron mass ratio $\mu =m_p/m_e\approx 1836$ is less visible because an atom responds only weakly to nuclear mass. Molecules vibrate, and their vibrational frequencies depend directly on nuclear masses.

The simplest molecule, HD⁺, contains one proton, one deuteron, and one electron. Two groups advanced its precision spectroscopy in 2020: Alighanbari, Giri, Constantin, Korobov, and Schiller (Nature 581: 152–158), and Patra, Germann, Karr, Koelemeij, and colleagues (Science 369: 1238). They determined μ at the part-per-trillion level while testing three-body QED. Kortunov and colleagues reached a resolving power of about 3 × 10¹¹, corresponding to a fractional transition-frequency uncertainty near 3.3 × 10⁻¹² (Nature Physics 17: 569, 2021).

Molecular hydrogen ions are therefore becoming an independent source of fundamental constants rather than only a test object. When two leading measurements of α already disagree at 5.4σ, independent routes matter.

Clocks that go to sea

In 2022, three optical clocks made by Vector Atomic sailed around the Hawaiian islands aboard a naval ship for twenty days. Each 35-litre package contained an iodine spectrometer, fibre-frequency comb, and controls. The clocks ran continuously and accumulated timing errors below 300 picoseconds per day (Roslund et al., Nature 628: 736–740, 24 April 2024).

This was not a precision record. The paper contrasts the sea clocks with laboratory optical clocks below 10⁻¹⁸ fractional inaccuracy, but the at-sea figure is a timing-error and instability result rather than a like-for-like accuracy comparison. The result mattered because leading laboratory clocks are not readily deployable: clock geodesy, dark-matter networks, and GPS-independent navigation are constrained by ruggedness as well as precision. sea deployment

Part IIthe nuclear clock grows up

From “we excited it” to “we constrained dark matter with it”

The main descent treated the thorium-229 nuclear clock as a striking demonstration: laser excitation of the nucleus at PTB in April 2024, then a direct comparison with a strontium clock in September 2024. The dark-matter application was a well-motivated hope, not yet a result.

A TU Wien-led collaboration now reports closing the feedback loop, the step that distinguishes spectroscopy from an operating clock. Continuous absorption spectroscopy supplied rapid feedback that stabilized a continuous-wave laser to the 148 nm transition of ²²⁹Th nuclei embedded in a millimetre-scale calcium fluoride crystal at room temperature. The group then searched the transition for periodic changes and slow drift over timescales from 20 seconds to one day.

The reported photon-coupling constraints are competitive with atomic-clock bounds. The reported sensitivity to the strong force and quark masses reaches parameter space electronic clocks cannot access because the thorium isomer energy is a near-cancellation of electromagnetic and strong-interaction contributions.

June 2026 preprint The result is arXiv:2606.04997, submitted 3 June 2026, and was not peer reviewed at the time represented here. If the result survives scrutiny, it establishes a new method. Its inferred reach also depends on an α-sensitivity factor known only to about 40%, $K=5900\pm 2300$ (Beeks et al., Nature Communications 16: 9147, 2025). Separately, precise nuclear sensitivity factors for the strong-force and quark couplings cannot yet be calculated, so the preprint shows illustrative benchmark bands at 10⁴ and 10⁵. That factor-of-ten span is a benchmark range, not a statistically quantified uncertainty distribution. This is precisely the Day 6 pattern: a tight experiment multiplied by a loose coefficient.

Part IIIthe clock as dark-matter detector

Sensors far apart, seeing different parts of the same wave

Suppose dark matter is an ultralight scalar field rather than particles striking a detector. If such a field couples to ordinary matter, constants could oscillate: electron mass would change slightly and clock frequencies would wobble.

For the lightest candidates, the wavelength can exceed a laboratory or even the solar system. Every local instrument then samples almost the same field at the same instant. Some couplings shift co-located references with the same response together and cancel from their comparison. For those reference pairs, the coupling is structurally invisible, not merely weakly constrained.

The 2025 solution was to stop making only local comparisons. Filzinger, Caddell, Jani, Steinel, Giani, Huntemann, and Roberts compared sensors separated in space and time so they sampled different field phases. Existing data supplied a frequency comparison between two optical cavities linked over 2,220 km, plus atomic clocks aboard GPS satellites. The analysis set the first constraints on scalar dark matter coupling to electrons alone, for masses from 10⁻¹⁹ to 2 × 10⁻¹⁵ eV/c² (Physical Review Letters 134: 031001, 23 January 2025).

separated-sensor method No dark matter was detected. The advance was a new window opened without a new detector: the geographical baseline became the apparatus.

Part IVgravity weighed by quantum things

Down to a flea’s egg—and then the interesting part starts

Gravity measurements usually use large source masses—Cavendish used lead balls, and modern measurements of G use kilograms of tungsten—because shrinking a same-density sphere rapidly weakens its signal. At comparable geometry the source mass scales as R³, so halving its radius costs a factor of eight in the available gravitational effect.

In 2021, Tobias Westphal, Hans Hepach, Jeremias Pfaff, and Markus Aspelmeyer measured gravitational coupling between two gold spheres of 1 mm radius and 90 mg mass, roughly four houseflies. A miniature torsion balance and 350 hours of integration reached systematic accuracy of 4 × 10⁻¹¹ m/s² (Nature 591: 225–228, 11 March 2021). It was the smallest source mass whose gravitational field had then been directly measured.

The Planck mass, about 22 micrograms, is a target scale in some quantum-gravity proposals rather than a hard experimental threshold. Reaching source masses around or below that scale could make quantum superposition of the source a plausible ambition, opening the question of whether its gravitational field must also be quantum.

From another direction, Holger Müller’s group replaced the couple of seconds available to free-falling atoms by holding them in an optical lattice. With an interferometer open for up to a minute, they measured a miniature source mass’s attraction as 33.3 ± 5.6(stat) ± 2.7(syst) nm/s², overall accuracy 6.2 nm/s²—more than four times better than the source’s free-fall comparison. The result agreed with Newtonian gravity and excluded screened-fifth-force chameleon and symmetron models over the natural parameter region analyzed (Panda et al., Nature 631: 515–520, 18 July 2024).

Part Vtwo routes to the ampere

Two ampere realizations—and an open triangle

The 2019 ampere is defined through elementary charge e: current is charge counted per second. The most literal realization is a single-electron pump, a nanoscale gate moving one electron at a time at known frequency f, so $I=ef$. At one gigahertz the current is about 160 picoamps—precisely defined but difficult to measure.

The target relative uncertainty for a practical direct electron-counting standard is roughly one part in 10⁸. Slow pumps produce currents too small to compare accurately; fast pumps miss or add electrons. The fixed-e definition therefore preceded a sufficiently accurate direct-counting realization, although the ampere could still be realized through other electrical standards.

The chronology matters. In 2017, a silicon single-trap pump exceeded 1 nanoampere at 7.4 GHz, but its current deviated from the ideal value by about 20 parts per million and was measured with about one-part-per-million uncertainty—well short of the 10⁻⁸ target. In 2025, a scalable silicon demonstration parallelized four pumps at 200 MHz and produced a plateau above 2 nanoamperes with three pumps at 2.1 GHz. That established a route to higher current, not 10⁻⁸ accuracy.

A separate route combines the Josephson voltage and quantum Hall resistance standards through Ohm’s law, using a superconducting cryogenic current comparator for scaling. A 2025 Nature Communications experiment generated microampere currents with relative uncertainties below 10⁻⁸. That is an established ampere realization, but it derives current from the Josephson and quantum Hall effects rather than independently counting electrons.

The quantum metrology triangle is the independent comparison: does current from a single-electron pump agree with current derived from Josephson volts and quantum Hall ohms through $V=IR$? A Josephson-plus-quantum-Hall generator alone reuses two sides and therefore cannot close the three-sided test.

If repeated independent triangle tests failed after their systematics were excluded, the consistency of the three quantum electrical effects—or of their realizations—would need revision.

The independent electron-pump comparison needed to close the triangle at 10⁻⁸ remains unfinished. PQCG ampere triangle closure

Part VIthe wildest proposal

Can you detect a single graviton?

Freeman Dyson argued that detecting one graviton was effectively hopeless: gravity couples too weakly, and a detector large enough to compensate might collapse under its own gravity or need to be the size of Jupiter.

In 2024, Germain Tobar, Sreenath Manikandan, Thomas Beitel, and Igor Pikovski proposed a different route (Nature Communications 15: 7229). The analogy is the photoelectric effect. Light’s quantization was inferred from discrete energy absorption rather than from visually catching a photon.

The proposal does not specify one universal multi-tonne niobium detector. The required mass and material depend on the target source: the paper estimates about 15 kg for a beryllium resonator tuned to GW170817 and also analyzes tonne-scale bars. The resonator would be cooled to its motional ground state at millikelvin temperature. When an astrophysical gravitational wave passes, a quantized interaction could drive one-phonon jumps corresponding to absorbed energy $E=\hslash \nu$. Continuous quantum-state monitoring would resolve the jumps and correlate them with LIGO’s observation of the same event: a gravito-phononic photoelectric effect.

single-graviton claim Nothing has been detected. The paper establishes a theoretical feasibility argument, not an experiment. No device combines the required resonator mass, quality factor, millikelvin temperature, and nondestructive single-phonon readout. An interpretive caveat remains even if discrete absorption were observed: a quantized absorber can receive energy in lumps from a coherent classical wave, so quantized transfer is not automatically identical to proving that the gravitational field itself is quantized. The proposal is conceptually striking and far from realization.

Part VIIwhen peer review is not enough

The Aziz–Howl dispute, and what it teaches

The setup was the proposed quantum-gravity witness: entangle two masses through gravity alone, and infer that gravity must be quantum because a local classical channel cannot create entanglement.

Aziz and Howl’s peer-reviewed Nature paper (646: 813–817, 22 October 2025) argued that in quantum field theory classical gravity can transmit quantum information and generate entanglement. If correct, the witness would no longer distinguish classical from quantum gravity.

Technical replies followed rapidly. Marletto, Oppenheim, Vedral, and Wilson argued that the model does not produce the claimed entanglement and that any remaining effect would be mediated by quantized matter. Diósi’s response was titled “No, classical gravity does not entangle quantized matter fields.” Di Biagio defended a route by which it might. Gundhi and colleagues argued in April 2026 that discarded transition amplitudes generate the apparent effect and that restoring them preserves an initially factorized state. Xue and colleagues also posted a preprint response. The rebuttals cited here remain preprints, and the dispute is active and contested, not resolved by counting papers.

Open questions

What would settle these

• Does classical gravity entangle? The dispute is over specific terms in a calculation. A peer-reviewed synthesis, convergent calculations, or agreement about the discarded amplitudes could resolve it.
• Can an HCI clock beat 10⁻¹⁸? Success would provide a much stronger α-sensitive instrument; irreducible motional systematics could stall the programme at the demonstration stage.
• Will an independent electron-pump leg close the quantum metrology triangle at 10⁻⁸? Failure after independent checks would be major evidence that a realization—or, only after systematics were excluded, the underlying physics—is incomplete.
• Can the thorium sensitivity factors be narrowed? The measured α-sensitivity factor K carries about 40% uncertainty, while strong-force and quark sensitivities are currently represented by illustrative benchmarks; nuclear-clock bounds inherit that theory dependence.
• Does antimatter fall at exactly g? The direction is known; fourteen orders of magnitude separate its present precision from ordinary-matter equivalence tests.
• Could a clock network ever detect an oscillating constant? A reproducible α oscillation matched across clocks and an inferred dark-matter frequency would overturn today’s null picture. No such signal exists.