# What is the Clock? — context

## The definition, and the one word that makes it work

The post defines a clock as a sequence of ticks, C = (t₀, t₁, t₂, ...). The author's first attempt was "a steady-state increasing counter." That bundles three separate properties (a rate bound, monotonicity, discreteness), and he dropped it. His second attempt, "at a philosophical level": "something that needs to provide ticks at a regular interval. If we have that, then we calibrate how these intervals relate to real time."

The version in the post came from Feynman: pick something periodic, then count it. **Count** is the bridge word. The author felt a jump between "something that repeats" and "a sequence of ordered events": "I jump to us having to label which happened before which... that's kind of like implicit, but it's also necessary. I'm not sure how to bridge that gap." What you do with a repeating thing is count it. Once that is said, indexing by the natural numbers is not a leap.

What the count leaves out is the point of the post. The label 3 says three sunrises have happened. It says nothing about how long that took, and makes no claim that the three days were equal. An early draft blurred this as "the difference between one event and the other need not tell us how long that took"; there is no difference object to point at — the count is the only thing the clock produces.

The paper this series leads to defines a clock differently: a function from real time to clock time. (Its model file was read directly: a hardware clock $H_u(t)$ with a bounded rate, and a logical clock $L_u(t)$ whose skew is bounded. The claim that this is also the standard textbook framing was stated from memory in the conversation and never checked against Lynch or Attiya–Welch — do not attribute it.) That is the special case the post is building toward, once a reference has been designated. The post deliberately does not open with it.

## Is a sundial a clock? (a distinction manufactured and withdrawn)

"A sundial is not a clock; the Earth is the clock and the sundial is the display" was offered to show the definition doing work. The author pushed: "you can represent a sundial as a count, right?... What is fundamentally different here?" The first defense — a clock counts, a sundial measures, "an odometer versus a ruler" — was withdrawn: too slippery, either can be redescribed as the other. The second — cover the device for two hours; the sundial is still right, a stopped pendulum clock is two hours wrong forever, so the sundial holds no state — is true, but the author still asked what a local oscillator gives that a sundial interface does not.

Final position: a sundial is a clock. Behind an interface it gives a monotone reading correlated with the Earth's rotation, and that is what a clock is. The useful question is never clock-or-not; it is the parameters. Resolution (you cannot build a sundial that resolves a microsecond). Availability (nothing at night or under cloud). And independence: two pendulum clocks side by side are genuinely two clocks and can disagree; two sundials cannot disagree about anything — they are one clock seen twice. "All the sundials on Earth are a single clock with many windows."

This is where Wikipedia's "measures and displays time" comes apart: two jobs, and the definition does not notice they are separable. The sundial displays; the Earth measures.

## Why the post attacks the Wikipedia definition, and the honest version of the objection

The post's stated objection is circularity: "measures and displays time" uses time to define a clock, when the series wants to use clocks to define time. The author worried this was weak: "I'm still using the word time... It's not like I'm doing it much different."

Two things were sorted out.

The circularity charge only bites on a reader who already grants that time is defined by clocks. A reader can say time is primitive and clocks merely measure it — Newton's position — and it is coherent. That reader is not trapped. The objection that needs no prior agreement: the definition names a purpose, not a mechanism, so it gives you nothing to build on. And the two definitions pick out different objects: a sundial displays time and contains no oscillator; a caesium standard contains an oscillator and displays nothing.

On "still using the word time": count how many times the word is needed to state each definition. Wikipedia's cannot begin without it. The post's needs it zero times — events, an order on them, a count. "Before" is a legitimate primitive, not smuggled-in time; the author's own field runs on it (Lamport's happened-before is a partial order on events with no physical clock in it).

Neither definition is wrong. Wikipedia's is the everyday one; oscillator-plus-counter is the physics and metrology one (it is what Taylor and Wheeler use when they build a clock from two mirrors and a photon). The difference is audience.

The post is careful not to claim it is defining time. An early draft said "a more abstract approach to defining time" four lines above quoting Feynman declining to define time. What the post defines is a way to count; post two adds a unit.

## Why regularity is a relation between two clocks, not a property of one

The author spent the most effort here, and readers push hardest here. Three arguments, in the order they became convincing.

**Nothing to be even against.** A man flipping an hourglass cannot tell whether the sand runs slow at night. Nothing in the device speaks to it. Feynman concedes exactly this and the post quotes him: "All we can say is that we find that a regularity of one kind fits together with a regularity of another kind." (The post's first draft had the hourglass sentence inverted — "how can he ensure that the hourglass is slow at night" — it must read "rule out.")

**Comparing two clocks needs no standard.** This was a sticking point: "What is the drift? Drift from what? A second? But we haven't even defined a second." What unstuck it: put two clocks side by side and calibrate them against each other — say a million ticks of one per turn of the other's hand. Watch the ratio. If it holds, they agree; if it moves, one is wandering, though nothing tells you which. No external standard, no definition of a second, only two counts. That is the entire content of "two clocks agree," and it is what the flips-against-days plot draws.

**The trap, and why it only partly closes.** The author set it on purpose: "How do we know that the atomic clock is not the one being too fast or too slow and not the day?... I think we do seem to hold a definition of regularity that's not in relation to another clock." The answer has three parts, and the first is the honest one.

Strictly, you cannot know. If every atomic clock sped up together, nothing would detect it. Feynman wrote "reason to believe," not "know."

"The Earth is irregular" means the Earth disagrees with a large cluster of clocks that agree with each other and share no mechanism: quartz, pendulums in vacuum tanks, caesium, hydrogen masers, optical clocks, millisecond pulsars. For the Earth to be the correct one, all of them would have to deviate together. Types matter: a room full of quartz alone would be weak evidence, because every crystal shares the same temperature sensitivity and temperature is seasonal — exactly the confound. Unrelated physics agreeing is what kills it.

And there is more than a vote. The Earth's deviations have independent causes that predict them in advance. The seasonal wobble comes from winds shifting angular momentum between atmosphere and ground and can be computed from weather data; tidal friction from the Moon slows the rotation and is checked against Babylonian eclipse records; the 2011 Tōhoku earthquake shortened the day by a computable amount. Nothing comparable explains a conspiracy among atoms. Plus an asymmetry the trap assumes away: there are thousands of caesium clocks and they can be checked against each other; there is one Earth, and it can never be compared to a twin.

Still a choice, not a proof. You could declare the Earth regular and every clock ever built wobbling in unison each year. Nothing forbids it. It is an appalling description — a coincidence among unrelated mechanisms, and you lose the ability to predict the wobble from the weather. That is Poincaré's "convenient," and it means something specific: the description under which the laws come out simplest.

## Straight at any slope: what the plot shows

Two errors were caught in the figure.

The first version had a dashed "equal rates" diagonal at slope 1. With 24 flips to a day the slope is 24, so slope 1 means nothing except a shared unit. Regularity is that the line is straight, at any slope. The slope is a conversion factor, the number you write on the ruler. Twenty-four flips a day and twenty-seven are equally regular; in the author's words, "if we have a straight line at all, we can just adjust to there." A wrong slope only becomes an error once somebody has designated a nominal slope.

The first published version had no irregular line. Both lines were straight, so "linearity indicates regularity" had nothing to contrast against and read as a definition rather than a distinction. The curved dotted line — rising fast, crossing the solid line, ending below it, so that no flips-per-day number can be written down for it — is the missing half of the argument. A reader (Henry) got the point anyway: "Ground truth is actually the stacking of relations on top of each other," and "what phenomena can we bundle together to form some invariant?"

Words for the figure: "plot," not "graph" or "image" (both carry mathematical meanings; a reader flagged "image"). The horizontal axis is a count of days and the vertical a count of flips; neither is real time. The lines are C′ and C″. In distributed algorithms the analogous picture is not a line but a band: the model bounds the instantaneous rate, so a clock may be any curve inside a cone around the ideal. The post's line is one actual clock; the field's cone is the region a clock is allowed to wander in.

## Why the post stops before skew, drift and offset

The author wanted to end post one by defining skew and drift, and could not. Every attempt collapsed.

Skew at day d was written C₁(d) − C₂(d). Then the parameter had to be freed from the day ("fix any clock and call its count t"). Then it needed a clause that the two clocks are "supposed to agree" — two hourglasses both built to run an hour, so a gap means something is off, unlike days versus flips, where nobody expects the numbers to match and only the ratio is meaningful. The author: "What does 'supposed to agree' mean?... The definition is still bad." Then the textbook definition was pulled in (|L_u(t) − L_v(t)| ≤ s, indexed by a real time nobody can read), which reversed the post's premise: "You just completely flipped the script dude. Our post is on the flow of showing how time is defined by clocks."

Resolution: cut the definition. Skew, drift and offset all require a designated reference. Straightness does not. So post one ends at regularity, and the designation is post two. The author landed on this three times before accepting it.

A second reason drift could not be defined here, which readers from either field will hit: "drift" means two things. In distributed algorithms it means the rate is wrong — a perfectly straight line with the wrong slope has drift, and that is what the paper's ϑ bounds. In metrology it means the rate is changing — the straight-but-wrong case is a frequency offset, and drift is reserved for the slope moving (what a hydrogen maser does, on the order of 10⁻¹⁶ per day). The series uses the distributed-algorithms sense once there is a reference. For the same reason the plot's curved line is not "drift"; it is Feynman's separate point that a bad choice of tick gives a bent line against the sand.

## The noon-to-noon confusion

The post says "count noon to noon." Readers ask why noon. The author asked the same and it took several tries.

Feynman (§5-2, verified): counting turnings of the glass "from each morning to the next," the number of hours "was not the same each 'day.' We should distrust the sun, or the glass, or both. After some thought, it might occur to us to count the 'hours' from noon to noon."

The author's objection: "If sunrise occurs two minutes earlier, then the next sunrise will roughly occur two minutes earlier. The difference between the two is still roughly [24 hours]." He was right, and the first explanation ("sunrise slides, noon does not") was wrong as stated. A constant shift is harmless: every sunrise-to-sunrise interval would be 23h58m, a straight line at a different slope, which by the post's own argument is fine.

What breaks it is that the shift is not constant and it reverses sign. Near the solstices sunrise barely moves for weeks; near the equinoxes it moves a minute or so a day; earlier from December to June, later from June to December. So flips per day come out 24, then about 24.03, then 24, then about 23.97 across the year — a slope that changes, a line that bends. Noon-to-noon stays at 24 the whole way around: daylight is a window that widens and narrows symmetrically about its middle; noon is the middle, pinned to the rotation, and sunrise is the edge. Noon is findable without a clock — the shadow of a stick is shortest then.

Three different things get called "day length." Daylight duration swings by hours across the year. Noon-to-noon (one turn relative to the Sun) varies by about half a minute across the year from tilt and orbit — expected, known since antiquity, and the reason a sundial can sit a quarter hour off mean time. Neither is the irregularity of the rotation itself, next section.

## How big the Earth's irregularity is (a number the author kept getting wrong)

Across three rehearsals the author said the day varies by "a minute," then "a few milliseconds," then "100 milliseconds." The seasonal variation in one turn of the Earth, timed by a star crossing the same line in the sky on successive nights, is about one millisecond. The number matters because it says how good clocks had to be before anyone could see it, which is why this was caught in the 1930s and not in 1800. (Recalled figure; see the unverified list.)

## What time is

"I honestly don't even know what time is. I think that's my problem." The answers that helped, in order.

Physics declines the question. It tells you what time does and how to measure it. Feynman: "What really matters anyway is not how we define time, but how we measure it."

"Regular" is circular — equal intervals presuppose time — and there is no definitional way out. The escape is empirical: many physically unrelated processes (a swinging weight, a vibrating crystal, an electron flipping in an atom, a decaying particle, a spinning collapsed star) keep step with each other to parts in 10¹⁵. Nothing licenses calling any one of them regular. What licenses the concept is that they all agree. Feynman again: "These 'best' clocks are those which we have reason to believe are accurate because they agree with each other."

Poincaré says the same with the philosophy attached: "We have not a direct intuition of simultaneity, nor of the equality of two durations. If we think we have this intuition, this is an illusion." The rules that replace it are chosen "not because they are true, but because they are the most convenient" — "the fruit of an unconscious opportunism."

A clock gives you a count, not an origin. Two perfect stopwatches started five minutes apart disagree forever; neither is wrong. "What time it is" has two independent parts: a unit, which is a recipe anyone can rerun, and a count from a chosen origin, which no experiment recovers — something has to have been counting the whole time. The author's objection: "how do we do carbon dating?" The sample is the clock; it has been decaying since the organism died, and Feynman (§5-4) calls radioactive material a "clock" for exactly this. It still cannot do timekeeping: precision about one part in a hundred, calibrated against tree rings, measures only time since one event.

Relativity, which the author was reading alongside (Taylor and Wheeler): proper time is the reading of one clock present at both events. It is the one number every frame computes identically, because reading one clock twice involves no synchronization between separated clocks. Taylor and Wheeler never give a one-line definition of a clock. Exercise 3-3: "we have made no distinction among spring clocks, quartz crystal clocks, biological clocks (aging), atomic clocks, radioactive clocks, and a clock in which the ticking element is a pulse of light bouncing back and forth between two mirrors." Exercise L-2 draws the post's line: a device that ticks unevenly is sold by a swindler; a good clock ticks evenly. Aging is in that list for a relativity reason (every clock must slow by the same factor), not because it is a good clock — it is good to within years.

## How time becomes standardized

The post ends: we cannot tell whether C′ is slow or C″ is fast; we must agree which to treat as correct and calibrate the other to it; nothing in the comparison forces the choice. What came up around that ending:

An earlier ending said "we must agree that a day is 24 hourglass flips." That designates the Earth — which is exactly what post two shows people abandoning — and it makes the choice sound obligatory. The published version keeps it a choice: fix the day, or fix the hourglass and let the day be the one that varies. Both are consistent with everything observed.

The tie is not broken by measuring more carefully or for longer; the comparison stays symmetric. It is broken from outside, by agreement, and an agreement about a clock is what a standard is. Poincaré's two watches (verified, vault scan p. 237): "there is not one way of measuring time more true than another; that which is generally adopted is only more convenient. Of two watches, we have no right to say that the one goes true, the other wrong; we can only say that it is advantageous to conform to the indications of the first." Those two watches are C′ and C″.

"Convenient" in Poincaré means simplicity of the physical laws, specifically. The ephemeris second (1960) is that move enacted: the Moon's motion looked irregular when timed by the Earth's rotation; redefining the second by the Earth's orbit made Newtonian gravitation clean again. Caesium was chosen for a different kind of convenience — engineering (a countable microwave frequency, one stable isotope, easy to vaporize). One word should not carry both.

The author's own line from rehearsal, which became the hinge of his talk: "Drift is measured against a standard. Skew is measured against agreement." Drift is checkable alone against a definition. Skew does not exist until there is somebody else. Offset is skew against the designated clock. All three arrive in post two.

The talk's thesis, which is the series' thesis: time is not read, it is agreed, at every level. The second is a standard somebody chose; UTC is an average nobody holds; his protocol is one more layer of the same thing.

## Vocabulary that blurred, and how it settled

**Regular / periodic / stable / accurate / fixed.** Five words were tried; the author: "you keep on using new words." Settled: regular means the ticks are evenly spaced, and the only question is over what span — a laptop crystal is regular over a second, a maser over a day, caesium over a month. "Stable" is not a separate property; it is regular over a longer span. Accurate means the rate is right in absolute terms, which needs a standard. Fixed is a property of a number that does not change: caesium's note is fixed, not regular. Regularity belongs to a process (the counted wave); the atom supplies permanence. "Periodic" was dropped because to the author it does not imply evenly spaced.

**Tick vs pulse.** A tick is what one clock emits for itself; a pulse, in the paper, is a coordinated event across nodes. Ticks were not renamed to pulses.

**Display vs measure.** Two jobs. The post needs only the second.

**Ratio vs difference.** Two clocks counting different things (days, flips) give a ratio, the slope. Two clocks built for the same job give a difference, skew. A day-counter and a flip-counter have no meaningful difference.

**Unit vs count.** The second is a recipe; the count since an origin is not.

**Definition vs machine** (post two territory). The definition describes an ideal atom nobody has; every real caesium clock is an attempt at it. Nothing is more accurate than the definition; plenty of things are more accurate than somebody's machine.

## Explanations that failed

Each lost the author and will lose a reader the same way.

"A clock counts, a sundial measures." Collapsed on the first push.

"Sunrise slides, noon does not." Wrong as stated; a constant slide is harmless.

A dashed diagonal at slope 1 labelled "equal rates." Wrong once the units differ.

Skew defined without a reference, patched with "kind of tick" and then "supposed to agree." Jargon for something that needed a designated clock.

Positing an unobservable timeline t "because nobody derives real time from clocks." Reverses the post's argument. The paper's t is a designated clock, idealized — bookkeeping for the person writing the proof, not a resource any node has.

The sealed-box slogan ("any clock in the box must agree with the light clock, or you could read the drift and announce your velocity"), repeated three times as a one-liner. Made no sense until unpacked: the drift is between two clocks inside the box, and no standard is involved.

Meta-phrases: "it quarantines the strangeness," "this is the real issue." "Does nothing for me."

Jargon lists ("Doppler from atomic motion, collisions, magnetic fields, finite interaction time"): "you cannot throw in these words." What replaced it and landed: the atom is a tuning fork, not a clock; it answers one pitch and ignores every other, like a wine glass that breaks at one note; what gets counted is a radio wave you generate, and the atom's only job is to say whether the wave is on pitch.

"One part in ten quadrillion." Costs a paragraph and nobody can feel it. "Sixteen decimal places" worked. "Two clocks would take 300 million years to drift a second apart" overreaches — it describes evaluated primary standards under evaluation, not clocks left alone.

Refuting a phrasing that was already right. "Is proper time the time the observer sees?" was answered "not sees, reads." The author: "You understand that's what I meant... I must then do double the translation here." The same happened with "the marks are consistently off" for the maser and with "measure the distance between the star and Mars"; each time a distinction was drawn that was not worth drawing, and each cost a turn.

Manufactured connections. The author asked whether the way labs synchronize clocks parallels his protocol and was first told the parallel was a deadline; he caught it ("those aren't pulses, that's just an event"). What survives is narrower: reading a shared reference beats measuring each other across a network.

## Explanations that landed

Count as the bridge from "periodic" to ℕ.

Two clocks side by side, watch the ratio. No standard needed, nothing tells you which is wrong.

The ruler with even marks and no numbers (a hydrogen maser): you can measure with it all day, you cannot say what a mark is worth without laying a real ruler beside it once, and the marks slowly widen together, so the number written on it goes stale. The author restated it in his own words — "a very consistent tick, but the ticks could all be consistently off" — and that version is the one to keep.

9,192,631,770 was counted, not chosen: how many caesium cycles fit in a second that already existed. "If anyone had been free to choose it, they would have picked ten billion. The ugliness is the fingerprint of a measurement." Like counting footsteps in a mile.

Scaffolding: the old second held the structure up while the new one was built, then came down. Today's definition mentions the Earth nowhere.

Only one Earth. It closes the loop on the post's own definition, and it was the line the author's talk landed on.

"The clocks caught the Earth being a bad clock" — a better beat than "the sundial got refined."

Stopwatches started five minutes apart: count without origin.

## Verified sources

Checked against a copy during the conversation. Wording from OCR scans should be confirmed before quoting.

**Feynman, *Lectures on Physics* Vol. I, Ch. 5**, from the author's vault copy (OCR; confirm at feynmanlectures.caltech.edu).
- §5-2: "What really matters anyway is not how we define time, but how we measure it. One way of measuring time is to utilize something which happens over and over again in a regular fashion—something which is periodic."
- §5-2: "We can just say that we base our definition of time on the repetition of some apparently periodic event."
- §5-2: the morning-to-morning versus noon-to-noon hourglass passage quoted above.
- Same chapter, the omnipotent being who slows the sand at night: "All we can say is that we find that a regularity of one kind fits together with a regularity of another kind."
- Same chapter, on the Earth's rotation: "These 'best' clocks are those which we have reason to believe are accurate because they agree with each other." (Section number not recorded.)
- §5-4: radioactive material as a "clock"; tree rings.
- §5-6: "Everybody knows that the way you measure distance is to start with a stick and count. Or start with a thumb and count. You begin with a unit and count." And on triangulation: "Although this means that we are really using a different definition of distance, when they can both be used they agree with each other."

**Poincaré, "The Measure of Time" (1898).** Read from the vault scan of *The Foundations of Science* (Halsted translation), in the part reprinting *The Value of Science*. Page numbers are for that scan: p. 237 (two watches), p. 243 (conclusion: "not because they are true, but because they are the most convenient"; "unconscious opportunism"). Primary publication: "La mesure du temps," *Revue de métaphysique et de morale* 6 (1898), pp. 1–13. A Modern Library edition exists (*The Value of Science: Essential Writings of Henri Poincaré*, ed. Gould, 2001, ISBN 9780375758485) but the page numbers above are not from it. Audoin and Guinot state that their Chapter 2 "is essentially a review of H. Poincaré's analysis."

**Taylor and Wheeler, *Spacetime Physics*.** Exercise 3-3 (the list of clocks) and Exercise L-2 (uneven ticks, swindler), from the author's copy.

**Audoin and Guinot, *The Measurement of Time*.** §3.2(c): "By definition, what we read on a clock C is the proper time t_C of this particular clock." §7.2 (TAI) at printed pp. 241–247; scan pages run about 12 ahead.

**Etymonline, "second"**: from Medieval Latin *secunda*, short for *secunda pars minuta*, "second diminished part," the result of the second division of the hour by sixty. (That word order is the source's; "pars minuta secunda" earlier was from memory.)

**Cristian 1989**, abstract as pasted by the author (the scan's "derivation" is an OCR error for "deviation"): "Externally synchronized clocks are also internally synchronized. The converse is not true."

**BIPM web pages**: TAI is computed "from a combination of data from about 450 atomic clocks operated by more than 80 timing centres"; Circular T covers "about 80 institutes." Other sources give 350 clocks and 69 institutes for other years; any figure should carry its date.

**The Aeneas thread on post one**, read directly: Henry's comments quoted above; the note that "image" has a mathematical meaning.

## Recalled, never confirmed

Stated from memory in the conversation and flagged as such at the time ("I am recalling all of this, not consulting anything"). Plausible; not established.

- Seasonal length-of-day variation ≈ 1 ms; caught by quartz clocks in the mid-1930s (Scheibe and Adelsberger, German standards lab). Winds and atmospheric angular momentum; tidal friction and Babylonian eclipse records; Tōhoku 2011 ≈ 1.8 µs.
- Solar day varies up to ~30 s across the year; the equation of time reaches about a quarter hour.
- Ephemeris second: 1/31,556,925.9747 of the tropical year 1900, from Newcomb's *Tables of the Sun* (1895); adopted 1960, replaced 1967. Tropical year ≈ 365.2422 days; sidereal year ≈ 365.256; sidereal day 23h 56m 04s. Hipparchus found precession by comparing equinox timings, ~130 BC.
- Essen and Parry (NPL) against Markowitz's lunar observations (USNO), 1955–58: 9,192,631,770 ± 20 cycles per ephemeris second; frozen into the definition in 1967.
- c defined exactly as 299,792,458 m/s in 1983, because laser measurements of c had become better than the metre standard (a krypton wavelength).
- Quartz at 32,768 Hz = 2¹⁵; a consumer crystal drifts ~10⁻⁵, about a second a day; commercial caesium beam ~10⁻¹³ (~9 ns/day); caesium fountain ~10⁻¹⁶; optical clocks ~10⁻¹⁸; redefinition of the second expected around 2030. Maser rate wanders ~10⁻¹⁶ per day. One caesium cycle ≈ 108 ps.
- Mechanical era: Huygens' pendulum (1656) reaching ~10 s/day within decades; Shortt free pendulum (1921) ~1 s/year, a sealed master pendulum that does no work and a slave that does all of it, corrected every 30 s — the same architecture as quartz disciplined by an atom.
- Hour, minute, second: 24 from Egyptian decan stars; 60 from Babylonian base-60 arithmetic; Ptolemy divided the degree by sixty twice; al-Biruni applied the scheme to time around the year 1000 (not verified in a primary source); "thirds" at 1/60 of a second (not verified); French decimal time decreed 1793, abandoned after ~17 months.
- Gravitational rate shift ≈ 1.1 × 10⁻¹⁶ per metre of altitude (≈ 9 ps/day/m); Boulder at 1650 m runs fast by ≈ 1.8 × 10⁻¹³, about 5 µs/year. GPS satellites net +38 µs/day (−7 velocity, +45 gravity).
- Books suggested: Galison, *Einstein's Clocks, Poincaré's Maps*; Landes, *Revolution in Time*; Sobel, *Longitude*; Jones, *Splitting the Second*; the BIPM SI Brochure for the record.

## Forward-looking: post two's territory

**Three designations.** The second was 1/86,400 of a day until clocks were steady enough to show the rotation irregular. Then a fraction of the Earth's orbit — the ephemeris second, steadier but nearly unusable, since pinning down a date that way took years of observation. Then, from 1967, 9,192,631,770 cycles of the radiation that flips caesium-133 between the two arrangements of its outer electron relative to its nucleus. The number matches the old second. Any atom would do — every atom of a kind has the same note everywhere, always — and caesium was the easiest to build around in the 1950s. The atom does not tick; it holds one pitch fixed, and a countable microwave from a quartz oscillator is locked to it. When someone says "atomic clock," the thing keeping time is the quartz; the atoms keep it honest.

**Why the Earth was dropped.** Not mainly because it wobbles. Because there is only one, so it cannot be checked against another Earth. Atoms come in unlimited identical copies. The post's own argument, applied.

**TAI and UTC.** TAI (Temps Atomique International — the French order stuck because the BIPM sits at Sèvres outside Paris, from the Metre Convention of 1875) is the count of SI seconds since 1958, maintained as a weighted average of roughly 450 clocks in about 80 institutes, computed monthly by the BIPM and published a month late in Circular T. Nobody holds it; no machine reads UTC in real time. UTC is TAI minus leap seconds (currently 37), kept within 0.9 s of the Earth's rotation so noon stays at noon; the CGPM resolved in 2022 to stop leap seconds by 2035. "UTC" is neither English (CUT) nor French (TUC), a deliberate compromise. Leap seconds track rotation; leap days track the orbit and the seasons — the author merged these once. Unix time counts from 1970, an unrelated choice.

**Unit and count in the infrastructure.** A dozen or so evaluated primary standards (caesium fountains) run in campaigns and report a rate with error bars; hundreds of commercial caesium clocks and hydrogen masers run continuously and carry the count without knowing their own rate. Steering ties the two.

**How distant labs compare** (common view). Neither sends the time. Both receive the same satellite broadcast; each records what its own clock read on arrival, subtracts its computed path delay, and they swap the numbers later by file. The satellite's error is in both readings and cancels; only path asymmetry survives, at the nanosecond level. Since 2006 the BIPM compares each lab against the whole GPS constellation (all-in-view). What transfers to the author's protocol: instead of two nodes measuring each other across the network at the cost of delay uncertainty, each consults a shared reference locally at the cost of how well it reads the reference.