Every Clock Agrees — the "Atomic Clock Analysis" paper, reviewed

01 · Provenance

What is this document?

By the author's own account this is the first paper he published and the longest — a thirty-two-page photo carousel on the Azimutha.FE TikTok account, titled on its cover card "Atomic Clock Analysis — Instrument Behavior vs Institutional Interpretation." It has a Preface, a glossary of "Terms and Definitions," six numbered sections, a Conclusion, and a two-column Summary, all under the running framework name "The Principle of Physical Consistency (PPC)." The caption promises "what nobody will tell you in reality." It exists only as screenshots — no hosted text, no named author, no checkable copy.

The argument is a single move repeated: an atomic clock counts atomic oscillations and converts the count to a time value by definition, its oscillation rate is sensitive to temperature, pressure, magnetic field, vibration and acceleration, and therefore when clocks read differently after being flown or put in orbit, that difference is "instrument variance" rather than time itself changing. Time dilation, on this account, is an interpretation layered onto instrument behaviour, propped up by "calibration circularity," withheld raw data, and one never-measured culprit the paper returns to: Earth's "toroidal" electromagnetic field.

One fact holds for every paper in this series and is worth stating plainly: the author reports no physical test of their own — no measurement, no apparatus, no data, no experiment they performed — only argument about other people's work. This paper indicts the most exhaustively characterised instruments in the history of measurement for insufficient measurement, while presenting none.

▸ Read the original paper in full, transcribed verbatim — his words, so you can check ours against them.

02 · Steelman

The strongest version of the paper's case

Its core technical facts are correct, and better stated than most. An atomic clock really does count oscillations — 9,192,631,770 cycles of the cesium-133 hyperfine transition define one second — and the conversion to "time" really is a definitional framework laid over a count. Those oscillators really are sensitive to temperature, magnetic and electric fields, pressure, and acceleration; characterising exactly those shifts is what precision metrology does all day. GPS satellite clocks really are corrected by about 38 microseconds per day. Hafele–Keating really was flown east and west only. Raw data and independent replication really do matter. None of that is wrong.

And one philosophical instinct is fair. There is no clock-independent "time meter": every measurement of time is some physical process standing in for time, so in a strict sense the instrument is always the only witness. A careful reviewer should grant that a single clock, read once, cannot by itself tell you whether "time" changed or the clock did. That is a real point, and the paper is right to press it. So we grant it in full — and then ask the one question that settles the matter, the question the paper never asks: when you build clocks on completely different physics, with completely different things that could go wrong, do they agree?

03 · The Central Move

"The instrument changed; whether time changed is a claim, not a demonstration" — and why every new clock defeats it

Start with the case that strips the paper's move to nothing: a muon is not a clock. It has no oscillator, no microwave cavity, no dial to misread — it is an unstable particle that simply decays. Fast muons decay slower, by precisely the relativistic factor. The paper's entire thesis is that the instrument changed — whether time changed, it says, is "a claim, not a demonstration." The muon is the case with no instrument, and it changes anyway. When the thing with nothing to disturb runs slow on schedule, there is nothing left to relabel as instrument variance. That is time. Everything that follows is the same result made general — the more kinds of clock you line up, the more the excuse runs out.

The whole paper rests on a gap it declares unbridgeable: a clock measures its own oscillation, not time, so a clock that reads differently proves only that the clock was disturbed. Stated for one clock in isolation, that is almost fair. It collapses the moment you have more than one kind of clock. Environmental disturbances are specific — a temperature swing, a stray magnetic field, a vibration each move a particular oscillator by a particular amount that depends on how that oscillator is built. A cesium microwave clock and a strontium optical clock have wildly different sensitivities to every item on the paper's list; they run at frequencies a hundred thousand times apart on utterly different transitions. If the shift the paper is explaining away were "instrument variance," these two clocks would disagree, in proportion to their different sensitivities. They agree — to parts in ten thousand trillion. And that is simply the definition of a valid measurement: when a dozen instruments with a dozen different failure modes all report the same number, that agreement is what measuring something real means. A disturbance that moved every differently-built instrument by exactly the same fraction is not a disturbance; it is the thing the instruments are honestly reporting.

This is the same error the account made about gravity and about the spin of the Earth, one level up. In the gravimeter paper every instrument's reading "could equally be electromagnetic"; in the gyroscope paper every device was "uniquely blind" to Earth's rotation. Each time the move is to grant that an instrument could be perturbed and then treat "could be" as "was — by exactly the amount needed to fake the result." But a possible confound is not a present one, and it certainly is not a confound that happens to equal the observed effect across a dozen unrelated machines. The more independent methods you line up, the worse the coincidence the paper needs. For its thesis to hold, a stack of physically unrelated error sources — cesium's magnetic sensitivity, strontium's light-shift, a muon's nothing — would all have to conspire to produce the identical relativistic number, over and over. That is the opposite of the simple explanation the paper claims to prefer.

And one fact closes the last escape: the environments do not have to differ at all. In 2010 two aluminium-ion clocks in one NIST lab were compared with one raised 33 centimetres above the other — same room, same air, same field — and the higher clock ticked faster by exactly the predicted gh/c². In 2022 a JILA strontium clock resolved the same shift across one millimetre inside a single cloud of atoms. There is no "different environment" a millimetre away; there is only height.

04 · Claim by Claim

The sections, audited

Each card: what's real in the paper's reasoning, the premise doing the work, the test that would tell, and the public record.

Section 1 · What an atomic clock "actually is" Misleading

"An atomic clock counts oscillations under controlled conditions — it measures its own behavior, not time itself"

Because the instrument produces a count that is converted to a time value by a human definition, it never measures time as an independent physical phenomenon; the conversion is a framework, not a measurement.

What's true

All of the description is right, and the standard model of physics agrees with it word for word: a clock measures the physical process of oscillation and calls the count "time." Nobody claims a clock detects a mystical time-fluid; a clock measures elapsed time along its own path, by counting a periodic process. The paper has correctly described what every physicist thinks a clock is.

The premise doing the work

That "counts oscillations, therefore doesn't measure time" is a distinction without a difference — and it proves too much. Every clock counts a physical process: a pendulum counts swings, a sundial counts the Sun's angle, the Earth counts its own rotations. If "it only counts its own behaviour" disqualifies a clock from measuring time, it disqualifies the sundial and the calendar the paper later calls real time in Section 5. The move quietly redefines "measuring time" to mean "measuring time without any physical process," which is measuring nothing. Time just is what accurate clocks agree on; the test of whether a count is a time measurement is whether independent counters agree, and they do.

The test that would tell

Build two clocks that count different processes — cesium's microwave hyperfine transition and strontium's optical transition — and see whether their "arbitrary human counts" track each other over months. They agree to better than one part in 10¹⁶. Two unrelated counters locked together are not each measuring "their own behaviour"; they are both measuring the same real thing.

The public record

The SI second is defined as 9,192,631,770 cesium-133 cycles specifically because that count is reproducible in any lab on Earth to extraordinary precision — the mark of a measurement of something real, not a private instrument quirk. Optical clocks on entirely different atoms now agree with it and each other far beyond the cesium definition's own precision.

Section 2 · The Hafele–Keating experiment Refuted by Data

"Only east and west were flown; the variables were never characterized — partial testing sold as full confirmation"

North–south flights were never conducted, and temperature, vibration, pressure and electromagnetic differences aboard the aircraft were never simultaneously measured and eliminated, so the reading difference cannot be attributed to time.

What's true

The history is accurate: two directions were flown, cabins are noisier and cooler and lower-pressure than a lab, and a single 1971 flight is not, by itself, an airtight isolation of every variable. Fair enough — which is why it is one experiment among many, not the whole case.

The premise doing the work

The "north–south gap" is a physics error. The altitude (gravitational) term does not care which compass heading you fly — it depends on height, not bearing — and the velocity term is a vector the theory already handles; a north–south flight introduces no mystery variable, just a different projection of ground speed the equations predict. And the deeper point: the predictions were published in advance, with error bars, and the flights matched them. Westward, the prediction was +275 ± 21 nanoseconds and the clocks returned +273 ± 7; eastward, −40 ± 23 predicted against −59 ± 10 observed. You cannot pre-register a signed, numbered prediction and then hit it if you are really measuring uncontrolled cabin noise — noise does not know which way the plane is flying, but the relativistic velocity term does, which is why east and west differ in sign.

The test that would tell

Repeat it with better clocks and the environmental terms measured, and see if the relativistic prediction keeps landing. It has, for fifty years — with hydrogen masers, optical clocks, and single instruments that never leave a fixed environment.

The public record

Alley's 1975–76 Maryland flights carried far better clocks and monitored the environment continuously; Vessot's 1976 Gravity Probe A flew a single hydrogen maser to 10,000 km and confirmed the gravitational shift to 70 parts per million — one instrument, up and back, so there is no "different clock" to blame. The 33-centimetre and 1-millimetre lab results close the loop with no flight at all.

Section 3 · GPS clock corrections Refuted by Data

"The 38-microsecond correction is an instrument correction, not a time correction" — with a cell-tower analogy insinuating GPS runs on ground-based line-of-sight

A correction that makes the system work does not identify what it corrects; instrument variance in orbit is the simpler cause, and the receive-only, line-of-sight, ground-station-anchored architecture is consistent with terrestrial transmission.

What's true

The plumbing is described correctly: civilian receivers are receive-only, the system uses ground monitor stations, and high-frequency signals do travel line-of-sight. And it is true in general that "a correction that works" does not by itself name its own cause.

The premise doing the work

Here the general point fails on the specifics, because this correction is not tuned to make things work — it is calculated in advance from relativity and built into the hardware before launch. Every GPS satellite clock is deliberately set to run slow, to 10.22999999543 MHz instead of 10.23 MHz, so that once relativity speeds it up in orbit it reads correctly. If the offset were empirical "instrument variance," it would drift with solar activity, season, satellite, and orbital position; instead it is the fixed number the formula predicts, ~45 microseconds a day gained from altitude minus ~7 lost from speed. Turn the correction off and position error grows by about ten kilometres per day — the theory's number is right to the microsecond, continuously, for decades. As for the cell-tower analogy's insinuation that GPS is really terrestrial: GPS works mid-ocean, over the poles, and on aircraft above any tower, and the satellites are tracked as orbiting objects by anyone with a receiver. The ground stations correct orbit and clock drift; they do not transmit your position from a mast.

The test that would tell

Ask whether the required correction is a free parameter or a prediction. If instrument variance, it would need continual empirical re-tuning per satellite; if relativity, one a-priori number works for all of them forever. The second is what operates.

The public record

Neil Ashby's Relativity in the Global Positioning System (Living Reviews in Relativity, 2003) documents the pre-launch frequency offset and the two relativistic terms as engineered-in, not fitted. GPS is the paper's strongest real-world case and it runs the other way: it is relativity working to the microsecond, every day, in a device in the reader's pocket.

Section 4 · Calibration circularity Not Demonstrated

"A clock calibrated within a relativistic framework cannot be independent evidence for relativity"

The second is defined within a framework that assumes relativity, GPS clocks are corrected with relativistic equations, and the clocks are then cited as proof of relativity — a closed loop.

What's true

Circular calibration is a real methodological sin, and worth watching for. If a measurement's calibration already presumed its conclusion, the measurement would indeed prove nothing.

The premise doing the work

The premise is simply false about how the second is defined. The definition — a fixed count of cesium cycles in the clock's own rest frame, at rest, at zero field — contains no relativistic assumption; it is a "proper time" standard, local to each clock. Relativity then predicts how two such independently-calibrated clocks will differ when one is moved. You calibrate each clock to the same atomic standard in its own frame, and then compare — the standard is frame-local, the prediction is about the comparison, and there is no loop. GPS does use the relativistic offset operationally, but that is applying a theory already validated by experiments — Hafele–Keating, Maryland, Gravity Probe A, the lab optical clocks — that assumed no such offset in advance. Using a confirmed theory to run a system is not circular; it is engineering.

The test that would tell

Point to the step where relativity enters the definition of the second. There isn't one: the count is fixed by a rest-frame atomic transition. The relativistic prediction is downstream, about what happens when clocks move relative to each other.

The public record

The BIPM definition of the second references the cesium hyperfine transition frequency of an atom at rest and at absolute zero — a local, framework-free proper-time standard. Every "confirmation" cited here is a comparison between such standards, predicted by a theory that had no vote in how each one was calibrated.

Section 4 · The raw-data problem Misleading

"Raw oscillation counts are rarely published; peer review is agreement within a closed institutional system"

Reviewers work from processed data inside the same framework and do not replicate the experiment, so the conclusions are institutional agreement, not independent physical demonstration.

What's true

Peer review is not replication, raw data is not always as accessible as it should be, and "many experts agreed" is not the same as "nature confirmed." Reasonable methodological hygiene.

The premise doing the work

The specific factual claim — that the environmental terms are uncharacterised and the data hidden — is the reverse of reality. Modern clock laboratories publish full uncertainty budgets: line by line, the blackbody-radiation (temperature) shift, the Zeeman (magnetic) shift, the Stark (electric-field) shift, gravitational and motional shifts, each with a measured magnitude and error, down to parts in 10¹⁸. The magnetic sensitivity the paper says was never measured is one of the most precisely quantified numbers in metrology. And the argument is fully general: "reviewers share a framework, so agreement proves nothing" would void every measurement ever made, including the sundial and calendar the paper endorses — those, too, are read within a shared framework about how the Sun behaves. An argument that disqualifies all measurement disqualifies none in particular.

The test that would tell

Open a modern optical-clock paper and read the systematics table. The NIST aluminium-ion clock's error budget (Brewer et al., 2019, Table I — freely readable) lists each shift on its own line — excess micromotion, secular motion, blackbody-radiation (temperature) Stark, quadratic Zeeman (magnetic), first-order Doppler, background-gas collisions — every one with a measured value and an uncertainty. If the shifts were uncharacterised, there would be no such table; there is, and it is the bulk of the paper.

The public record

The NIST aluminium-ion clock (Brewer et al., 2019) reports a total systematic uncertainty below 10⁻¹⁸ with every environmental shift itemised and bounded in its Table I; PTB and NPL publish the same for their strontium and ytterbium clocks. The "withheld raw data" is a published error budget anyone can open and read.

Section 5 · Time as a "natural" phenomenon Self-Contradicted

"Real time comes from observed natural cycles — the Sun, the Moon, the year — while time dilation departs from direct observation"

Time derived from the day, the ~28-day month, and the ~13-lunar-cycle year is directly observed and real; the claim that time itself stretches is a departure never demonstrated apart from the instruments.

What's true

The narrow point is correct and undisputed: time zones are a labelling convention, not time dilation — solar noon in Tokyo and in London is the same phenomenon at different longitudes. No physicist claims otherwise. It is a tidy refutation of a claim no one makes.

The contradiction

Section 1 disqualified the atomic clock for "only counting its own physical behaviour." But the Sun-clock and the calendar are the same kind of thing — the Earth's rotation is a mechanical oscillator counting its own turns, exactly as much an "instrument" as a cesium cell. The paper trusts the clock it never characterised and distrusts the clock characterised to eighteen digits. Worse, the Earth is the less reliable clock: its rotation is measurably irregular — earthquakes, glacial rebound and atmospheric angular momentum change the length of day — which is precisely why the world abandoned astronomical time for atomic time in 1967 and now inserts leap seconds to keep the wobbly Earth-clock in step with the steady atomic one. And the "directly observed" cycles are misstated: the lunar (synodic) month is about 29.5 days, not 28, and there are about 12.4 lunations in a year, not 13 — factual errors in the passage offered as unimpeachable direct observation.

The test that would tell

Compare the "real" astronomical clock against the "fake" atomic one over years. The atomic clock wins so decisively that we correct the Earth to it, not the reverse — the observable record of every leap second.

The public record

The IERS monitors Earth's variable rotation and announces leap seconds; the 1967 SI redefinition moved the second from a fraction of the year to the cesium transition precisely because the astronomical clock was too irregular to define time. The cycle the paper calls bedrock is the one metrology had to stop trusting.

05 · The Decisive Section

Common mode: the toroidal field

There is one intellectually serious version of the paper's argument, and it deserves to be named out loud rather than left implicit. It is not "each clock is disturbed" — we have seen that unrelated clocks agree. It is the fallback: maybe they share a common disturbance. The paper supplies the candidate by name in Section 6 — Earth's "toroidal" electromagnetic field, which does vary with altitude, and which (the reasoning goes) could nudge every electromagnetic oscillator the same way at the same height, faking a universal shift. This is the real objection, and the honest reason the convergence argument is not yet complete: convergence only proves a shared cause if the methods truly share nothing. So we have to show the shift survives when magnetism specifically is ruled out — not by trusting a metrologist's subtraction, which the paper will call circular, but by cases where a magnetic cause is excluded by construction.

Turn his own variable up thirty-thousand-fold and watch it do nothing. In the CERN muon storage-ring experiment the muons circulate through a magnetic field of about 1.5 tesla — some thirty thousand times stronger than Earth's ~50-microtesla surface field, and far beyond anything a toroidal-field gradient could contribute at altitude. If magnetism affected the decay, that field would obliterate the result. Instead the muons' lifetime is dilated by exactly the relativistic factor for their speed (a Lorentz factor near 29), matching special relativity to about a tenth of a percent. The strongest magnet in the entire subject produces the pure relativistic number, and the decay it acts on is a weak-interaction process with no oscillator for a field to detune. The paper's mechanism, cranked to thirty thousand times Earth strength, does nothing.

Remove the "up there" entirely. The toroidal argument needs the field to be different at a different height. The 2022 JILA result resolved the gravitational shift across one millimetre inside a single, actively shielded, actively measured atom cloud — one field, identical on every atom to far better than the effect. There is no "higher" for the field to differ at; the two ends of the sample sit in the same field a millimetre apart, and the shift still appears, scaling with that millimetre of height. You cannot blame a field gradient that is provably absent.

Let clocks with different magnetic sensitivities vote. This is the direct answer to the common-mode objection on its own terms. Aluminium-ion, strontium and ytterbium optical clocks have deliberately different magnetic sensitivities — the aluminium-ion clock was built to be almost immune. If a shared magnetic field were producing the shift, these clocks would disagree in proportion to their sensitivities. They agree — the Boulder network measured the frequency ratios between aluminium-ion, ytterbium and strontium clocks to eighteen digits. For the common mode to survive, a magnetic cause would have to move a magnetically-immune clock by the same fraction as a magnetically-sensitive one — a contradiction in terms — and this needs no subtraction to see: it is a raw comparison of dials engineered to respond differently, landing on one number.

And the shift has the wrong shape and the wrong sign for magnetism. A magnetic (Zeeman) shift scales as the field squared and tracks the field's profile; the relativistic shift is linear in height and independent of compass bearing, and its velocity term flips sign east-versus-west with the Earth's rotation. These are different functions of different variables, and the data fit the relativistic one. Two of the cleanest confirmations do not involve an atomic-clock transition at all, so there is nothing for the paper to claim was mis-subtracted: Pound and Rebka (1959) measured the gravitational shift of a gamma ray up a 22.5-metre tower using Mössbauer nuclei, and Ives and Stilwell (1938) measured the velocity time-dilation of moving hydrogen ions' light decades before atomic clocks existed.

The four things magnetism would have to do at once

To keep the toroidal-field line alive, Earth's field would have to: (1) do nothing to a muon at 1.5 tesla but everything to a clock at 50 microtesla; (2) produce a differential across one millimetre of a single, uniform, shielded field; (3) move a magnetically-immune aluminium-ion clock by the same fraction as a magnetically-sensitive one; and (4) reproduce the linear-in-height, sign-flipping form of gh/c² and v²/2c² while a magnetic shift goes as B². That is not one hidden variable. It is four mutually contradictory ones, invoked because the alternative is that the clocks are telling the truth.

And the same self-contradictions as the other six papers

The paper demands raw data and physical demonstration while performing no physical test of its own — the standing pattern of the series, at its sharpest here against the most-measured instruments ever built. It rests its whole conclusion on an unfalsifiable culprit: the toroidal field is never assigned a magnitude, never used to predict a number, only asserted to be "not demonstrated negligible" — which is exactly the move the author's own Kepler paper denounces as science's cardinal sin, the ad-hoc rescue that explains anything and forbids nothing ("the planets just chose to retrograde"). Here it is "an uncharacterised field of unspecified size did it." He built a paper attacking that structure and a paper standing on it. And Section 5 trusts the Earth's rotation as "real time" one page after Section 1 disqualified any instrument that "counts its own behaviour" — while the Earth-clock is the one we had to stop trusting in 1967 for being too irregular. The through-line of all seven prior papers surfaces again: a real sensitivity is granted, then silently promoted to the exact size needed to erase the result, and the burden of proof is placed on everyone except the author.

06 · What The Paper Never Mentions

The clocks that don't share the paper's excuse

A paper arguing that clock differences are cesium-oscillation artifacts omits every confirmation that uses no cesium, no microwave cavity, and in one case no instrument at all. Each is checkable.

The muon that isn't a clock: Cosmic-ray muons reach the ground only because their lifetime is dilated in flight (Rossi–Hall, 1941); in the CERN and Fermilab storage rings muons at 0.9994c live ~29 times longer, matching special relativity to ~0.1% — in a 1.5-tesla field. No oscillator, no dial, nothing for a field to detune. The particle simply lasts longer when it moves.
Optical clocks agreeing with cesium: Strontium, ytterbium and aluminium-ion clocks run on optical transitions a hundred thousand times higher than cesium's, with different sensitivities to every environmental term — and they agree with cesium and each other far beyond the cesium second's own precision. Unrelated physics, one reading.
The shift seen across 33 cm and 1 mm: NIST 2010: two aluminium-ion clocks, one raised 33 cm, same lab — the higher runs fast by exactly gh/c². JILA 2022: the same across 1 mm inside one atom cloud. No "different environment" a millimetre away; only height. The paper's confounds have nowhere to hide.
Gravity Probe A — one instrument, up and back: Vessot's 1976 hydrogen maser flew to 10,000 km and returned, its rate compared against its own ground twin, confirming the gravitational shift to 70 ppm. A single clock going up and coming down removes the "different instrument" excuse entirely.
GPS failing by 10 km a day without it: The relativistic correction is calculated before launch and built into the satellite clock frequency; switch it off and navigation drifts ~10 km per day. A theory that is merely "an interpretation" does not keep a global utility accurate to the microsecond for thirty years.
The published error budgets: Every modern optical-clock paper itemises the blackbody, Zeeman, Stark, gravitational and motional shifts to parts in 10¹⁸. The "uncharacterised variables" and "withheld raw data" the paper needs are a printed table in each of those papers.
That the Earth-clock is the wobbly one: The astronomical time the paper calls "real" drifts enough that we abandoned it for atomic time in 1967 and add leap seconds to keep it aligned. The cycle offered as bedrock is the one metrology stopped trusting.

07 · Anticipated Responses

Objections, answered in advance

"You characterised the shifts using relativity — the whole thing is circular.": Three of the load-bearing results have nothing to subtract. The muon lifetime and the Pound–Rebka gamma-ray shift use no atomic-clock transition, so there is no relativistic "correction" applied to them — the raw result is the effect. And the agreement between clocks built with different magnetic and light-shift sensitivities is model-independent: two dials engineered to respond differently, landing on one number, requires no shared subtraction to interpret. The definition of the second is itself framework-free — a rest-frame cesium count — so relativity enters only as a downstream prediction about moving clocks, never in the calibration.
"The toroidal field was never measured and could still be the cause.": It was tested at thirty thousand times Earth's strength — the 1.5-tesla muon ring — and did nothing to the decay. It cannot produce a shift across a uniform field one millimetre wide. It cannot move a magnetically-immune aluminium-ion clock as much as a sensitive strontium one, yet they agree. And its predicted signature (field-squared, tracking the field profile) is the wrong function of the wrong variable for a shift that goes linearly with height and flips sign with flight direction. A cause that fails all four tests is not an unexamined possibility; it is an excluded one.
"Predictive success isn't physical truth — you're confusing a working correction with reality.": Granted in general, and answered in the specifics. GPS does not fit its correction to make things work; it computes the number from relativity before launch and the hardware obeys it for thirty years. Hafele–Keating pre-registered signed, numbered predictions and hit them. A muon has no "model" to fit — it just decays slower. When independent methods predict unseen results in advance and land, that is the line between curve-fitting and physics, and every one of these is on the physics side.
"Real time is the Sun and the Moon, not a machine in a lab.": The Sun-and-Moon clock is a machine too — the Earth counting its own rotations — and it is the less reliable one. Its rate wanders with earthquakes, tides and the atmosphere, which is exactly why the world redefined the second onto cesium in 1967 and corrects the Earth to the atomic clock with leap seconds, never the reverse. The paper's trusted clock is the one metrology had to demote.
"The review didn't address the paper in its entirety.": The Preface, all six sections — the atomic clock, Hafele–Keating, GPS, calibration circularity, the raw-data argument, natural cycles, and the toroidal field — the Conclusion and the Summary are all addressed above, along with the caption. The standing invitation is unchanged: name one specific error on this page and it will be corrected, visibly.

08 · Primary Sources

Check us

BIPM — the SI second — defined as 9,192,631,770 periods of the cesium-133 ground-state hyperfine transition, specified for an atom at rest at a temperature of 0 K: a local, framework-free proper-time standard. The relativistic prediction is downstream, about comparisons between such standards.
Hafele, J. C. & Keating, R. E. (1972) — "Around-the-World Atomic Clocks: Predicted / Observed Relativistic Time Gains," Science 177, 166–170. Predictions published in advance: westward +275 ± 21 ns (observed +273 ± 7), eastward −40 ± 23 ns (observed −59 ± 10).
Bailey, J. et al. (1977) — "Measurements of relativistic time dilatation for positive and negative muons in a circular orbit," Nature 268, 301. CERN muon storage ring, Lorentz factor ≈ 29.3, field ≈ 1.5 T (~30,000× Earth's surface field); lifetime dilation confirmed to ~10⁻³. Final report (freely readable, documents the ring, its ~1.5 T field, and the time-dilation test): Bailey et al., Nucl. Phys. B 150, 1 (1979), CERN Document Server record 133132.
Chou, C. W., Hume, D. B., Rosenband, T. & Wineland, D. J. (2010) — "Optical Clocks and Relativity," Science 329, 1630. Two ²⁷Al⁺ clocks; time dilation seen at a 33 cm height difference and at ~10 m/s.
Bothwell, T. et al. (2022) — "Resolving the gravitational redshift across a millimetre-scale atomic sample," Nature 602, 420 (open-access preprint, arXiv:2109.12238). JILA strontium clock; gravitational shift resolved across 1 mm within a single sample — one field, no "different environment" to blame.
BACON Collaboration (2021) — "Frequency ratio measurements at 18-digit accuracy using an optical clock network," Nature 591, 564 (open-access preprint, arXiv:2005.14694). The Boulder network compared aluminium-ion, ytterbium and strontium clocks — three different atoms with different magnetic and light-shift sensitivities — and found their frequency ratios agree to parts in 10¹⁸: the direct answer to the common-mode objection.
Vessot, R. F. C., Levine, M. W. et al. (1980) — "Test of Relativistic Gravitation with a Space-Borne Hydrogen Maser," Phys. Rev. Lett. 45, 2081. Gravity Probe A; a single maser flown to ~10,000 km, gravitational shift confirmed to 70 ppm.
Pound, R. V. & Rebka, G. A. (1959–60) — "Gravitational Red-Shift in Nuclear Resonance" / "Apparent Weight of Photons," Phys. Rev. Lett. 4, 337. Gamma-ray (Fe-57 Mössbauer) gravitational shift up a 22.5 m tower — no atomic-clock transition involved. Refined to ~1% by Pound & Snider (1965).
Ives, H. E. & Stilwell, G. R. (1938) — "An Experimental Study of the Rate of a Moving Atomic Clock," J. Opt. Soc. Am. 28, 215. Transverse Doppler (velocity time-dilation) in moving hydrogen ions — the first laboratory confirmation, predating atomic clocks.
Ashby, N. (2003) — "Relativity in the Global Positioning System," Living Reviews in Relativity 6, 1. The ~38 μs/day correction (~45 gravitational − ~7 velocity), the pre-launch satellite-clock offset to 10.22999999543 MHz, and the ~10 km/day drift without it.
Brewer, S. M. et al. (2019) — "²⁷Al⁺ Quantum-Logic Clock with a Systematic Uncertainty below 10⁻¹⁸," Phys. Rev. Lett. 123, 033201 (open-access preprint, arXiv:1902.07694). Table I is a published, itemised error budget — blackbody, Zeeman, Stark and motional shifts each measured and bounded; the aluminium-ion clock's near-immunity to magnetic fields.
IERS / 1967 SI redefinition — Earth's rotation is measurably irregular (length-of-day variations from geophysical and atmospheric effects); the second was redefined from an astronomical fraction to the cesium transition in 1967, and leap seconds correct civil time to atomic time — the astronomical clock aligned to the atomic one, never the reverse.
Rossi, B. & Hall, D. B. (1941) — muon flux versus altitude: cosmic-ray muons survive to sea level only because their decay is time-dilated in flight; an early, non-laboratory confirmation using a particle lifetime rather than a clock.