The shape of the Earth, in plain sight.

• 1Eratosthenes (~240 BCE). The Sun was overhead at Syene at solstice noon (no shadow down a well); a gnomon at Alexandria ~800 km north cast a 7.2° shadow = 1/50 of a circle → circumference ≈ 50 × baseline ≈ 250,000 stadia. Depending on the stadion, that's ~39,000–46,000 km against the true 40,075 km — right the first time, with sticks and shadows.
• 2The three-point killer. Flat-earthers re-explain the shadows with a near Sun. But add a third and fourth station at other latitudes: the near-Sun model demands a different Sun height for each pair (a contradiction), while a distant Sun + spherical Earth fits all of them at once. The Sun's angular size also stays ~0.5° all day everywhere — impossible for a Sun supposedly approaching and receding overhead.
• 3The horizon dips. From height h the horizon sits below true horizontal by θ ≈ √(2h/R) — about 1.6° from a 1,000 m peak, measurable with a theodolite, and it dips further the higher you go. On an infinite plane the horizon would stay at eye level. It doesn't.
• 4The horizon recedes. Its distance grows as √height (≈3.57·√h km) — a finite, height-dependent edge, exactly a sphere and not a vanishing point on a plane. Surveyors, gunners, and microwave engineers all design for the drop (≈8 in × mi²).

Sources: Eratosthenes' measurement [43] · horizon geometry & refraction [8] · circumference [4]. → Curvature data rows

• 1The original error. Samuel Rowbotham (1838) sighted along the Old Bedford River just above the surface and saw a distant marker, declaring the water flat. Two faults: a near-surface sightline runs through the strongest temperature/refraction gradient (bending light down along the curve), and a single line of sight cannot separate "flat" from "curved + refraction."
• 2Wallace's fix (1870). Alfred Russel Wallace set three markers at equal height over six miles. On a flat surface they'd line up; over a curve the middle one rides higher than a line joining the ends. It did — by about the predicted amount — winning the £500 "Bedford wager." (John Hampden never accepted the result and harassed Wallace for years — a cautionary tale of belief versus evidence.)
• 3Why three points matter. The middle-marker bump is the signature of a bulge and is robust to uniform refraction — the same logic as the curvature calculator in Entry 3. Modern surveys with proper targets and a measured refraction coefficient reproduce it.
• 4It's an optics lesson. Graze the water and refraction can fake flatness; raise the sightline and use multiple targets and the curve reappears. Rowbotham didn't disprove curvature — he discovered refraction without realizing it. (See the Optics section.)

Sources: Bedford Level — Rowbotham & Wallace [44] · refraction [8]. → Curvature data rows

• 1Geometry — the hidden base. Past the horizon (~4.7 km for a 1.7-m eye, farther with height), the bottom of a target is occluded by the bulge of water or land. A zoom sharpens what is visible but cannot raise a base that is geometrically behind the curve. The "sinking ship / missing foundation" is the signature.
• 2Refraction — looming. Temperature gradients, especially over water, bend light downward and can lift hidden objects partly back into view (superior mirage). This explains many "impossible" shots — and it is a curvature-dependent atmospheric effect, not evidence of a plane. (See the Optics section.)
• 3Resolution — what a zoom does. More focal length resolves finer detail; it does not change line-of-sight geometry. "Zoom in and it comes back" restores detail that was still above the horizon — never a base that is below it.

The record — 443 km, and why it was perfectly possible

The longest confirmed line-of-sight photograph (Guinness World Record) is Marc Bret's 2016 image of Pic Gaspard in the French Alps (3,883 m) shot from Pic de Finestrelles in the Spanish Pyrenees (2,826 m) — 443 km apart. It is often cited as proof of no curvature. It is the opposite: the shot is only possible because of the curve, and it shows it.

• ①Altitude buys horizon. Horizon distance grows as the square root of height: d ≈ 3.86·√h km (with standard refraction). A 2,826 m peak sees ~205 km to its horizon; a 3,883 m peak sees ~240 km. Their horizons meet at ~445 km — essentially the 443 km separation. The two summits are just barely in mutual line of sight over the bulge.
• ②The photo shows the curve. It is not the whole Alps — only a thin row of summits rising over the horizon. Run the geometry and roughly the lower ~2,800 m of the Écrins massif is hidden behind Earth's curve; only the top tens of metres clear it. That hidden base is the curvature, recorded in the frame.
• ③Right at the limit. Because the summits sit within tens of metres of the threshold, the shot needed the exceptional refraction and clarity reported that dawn (a polarizing filter cut haze) — which is exactly why it's a record and not a routine snapshot. On a flat Earth there would be no horizon to clear, no hidden base, and far more distant low-lying objects would be photographed all the time.

Interactive — how much is hidden by the curve?

Enter an observer height, a target height, and a distance. The model returns the horizon distance, how much of the target is hidden behind the bulge, and how much clears it. Adjust the refraction k (0 = vacuum, 0.13 = standard air, ~0.2 = strong/ducting) to see how sensitive long shots are to the atmosphere.

Infrared photography

Longer wavelengths scatter less — Rayleigh scattering falls as 1/λ⁴, so near-infrared (~850 nm) scatters roughly 6–16× less than blue light. That's why landscape and long-distance photographers reach for red or near-IR filters: they punch through atmospheric haze and deliver crisp distant detail that visible light loses to scatter. But IR helps with haze, not geometry. The horizon is set by the curve, not the wavelength. Infrared makes an object that is already above the horizon clearer; it cannot image a base hidden behind the bulge. Long-distance IR shots show the same sinking-ship, hidden-foundation signature as visible light — just with less haze. If anything, IR refracts marginally less than visible light, so it lifts hidden objects slightly less, not more.

Still queued for this entry: observer-height tables and worked target curves (Toronto skyline, lighthouse nominal ranges) layered onto the calculator above.

Sources: horizon geometry & refraction [8] · longest-line-of-sight record [41] · atmospheric scattering & infrared [42]. → Photography rows (Optics)

• 1The dip has a formula. The horizon sits below true horizontal by θ = arccos(R/(R+h)) ≈ √(2h/R). That is ~0.045° at standing eye height (2 m), ~1° from a 1,000 m hill, and ~3° (about 2.98° once refraction is included) at a jet’s 33,000 ft. A flat Earth predicts 0° at every height.
• 2You can measure it yourself. Sailors have applied a “dip of the horizon” correction in celestial navigation for centuries; even at the shoreline it is ~2–3 arcminutes, within naked-eye resolution. From a plane window, apps like Theodolite or Dioptra show the horizon sitting clearly below eye level — no special gear needed.
• 3“Rises to eye level” is an illusion of no reference. Without an instrument the brain treats the horizon as level, because the dip is small and there is nothing to compare it against. Add a true horizontal and the flat prediction (0° dip, always) breaks — while the globe’s prediction lands on the numbers.

Sources: horizon-dip geometry & the navigation correction [96]. Builds on the curve of Entry 1 and the long-range sightings of Entry 3. → Curvature data rows.

• 1“8 in × mi²” is a parabola — and the wrong quantity. It is the small-angle approximation of the surface’s drop below a level line from your eye (the exact drop is R(sec − 1)); it stays accurate to a fraction of a percent out to a few hundred km, but it measures a tangent drop, not what is hidden — a tall, distant object can still poke above the bulge. This calculator skips the shortcut and uses the exact spherical geometry: a target an angle β past your horizon is hidden by R′(sec β − 1).
• 2Eye height moves the horizon — a lot. With your eyes at 2 m, the horizon is ~5 km away; from a 100 m hill it is ~36 km. Every metre of observer height lets you see farther and hides less of what lies beyond, which is why “but I can still see it” rarely matches the naïve drop.
• 3Refraction bends the ruler. Air thins with height, so light curves gently downward, effectively flattening the Earth for a sightline — the standard correction uses an effective radius about 7/6 of the real one. The toggle shows how much that alone changes the result. None of this is special pleading; it is the geometry surveyors and mariners have used for centuries (Entry 4, Entry 64).

Sources: horizon & hidden-height geometry [96]. Try the matching cases in Entry 2 (Bedford) and Entry 64 (the Causeway & salt flats). → Curvature data rows.

• 1"Level" is not "flat." Still water settles perpendicular to local gravity — an equipotential surface. On a sphere that surface is curved (the geoid). Water "finding its level" is exactly what a curved ocean does.
• 2The meniscus is tiny. Surface tension shapes water only within a capillary length (~2.7 mm). A meniscus is liquid climbing a container wall; scale the container up to a pond and gravity flattens the surface to the local equipotential. It never bows up across open water.
• 3Sea level genuinely curves. Satellite altimetry shows the geoid deviating roughly −106 to +85 m from a smooth ellipsoid, tracking gravity (mass distribution) — yet remaining globally curved. The drop is ~8 in × (miles)².
• 4Water is never truly "flat." Its shape is set by whichever force wins. At small scales surface tension takes over and pulls water into spheres — raindrops, beads of dew, the floating blobs astronauts play with in orbit. At planet scale gravity wins and the surface settles onto a curved equipotential; twice a day the Moon lifts the whole ocean into two tidal bulges (Entry 15). A calm lake only looks flat because the curve is gentle across a few miles — not because water can't hold a curve.
• 5Why the deep ocean is cold. The seafloor sits atop ~2,900 km of mantle and crust, and rock is a poor conductor (~2–3 W/m·K). Geothermal flux (~0.087 W/m²) is about 3,900× smaller than mean solar input (~340 W/m²). The ocean is warmed from above; cold, dense water sinks at the poles and fills the abyss at 0–4 °C. This says nothing about Earth's shape — but it's a common "gotcha," so it's answered.

Sources: geoid / sea level [28] · surface tension & capillary length [29] · ocean circulation & deep temperature [30] · geothermal vs solar flux [31]. → Water data rows

• 1The shadow that maps the core. S-waves (which can't travel through liquid) disappear everywhere past ~103° from the epicentre — revealing a liquid outer core starting at the Gutenberg discontinuity, ~2,890 km down. P-waves refract at that boundary, leaving a ring with no direct arrivals between ~103° and ~142°. The size and shape of those shadows fix the core radius (~3,480 km) and the planet's radius. The independent variable is angular distance from the quake; the dependent variable is which waves arrive when.
• 2It only works on a sphere. The shadow zones are symmetric rings at fixed angular distances no matter where on Earth the quake is — exactly what a sphere with concentric shells predicts. A flat disc of any thickness gives no such universal angle, and no reason for S-waves to cut off at one consistent arc. Inge Lehmann even found the solid inner core (1936) from faint P-waves sneaking into the shadow.
• 3Anyone can check the data. Global seismic networks publish arrival times openly; the PREM reference model (1981) was built from thousands of quakes and predicts travel times to within seconds. It is reproducible, falsifiable, and impossible to stage — the "experiment" is run for free every time the Earth shakes (compare Eratosthenes' surface measurement, Entry 1).

Sources: seismic shadow zone & Earth's layered interior (Gutenberg, Lehmann, PREM) [72]. Complements the surface measurement of Entry 1 and the gravity of Entry 9. → interior data rows

• 1Each tower is plumb, so they can’t be parallel. A tower built straight up points at Earth’s center. Two such towers 1,298 m apart are splayed slightly outward; over the 211 m height of the Verrazzano towers that opens a 41.3 mm gap at the top. It was an explicit design criterion in 1964, documented by the bridge authority and its engineers.
• 2You can weigh the curve with a tape measure. From the tower height (211 m), the spacing (1,298 m) and the 41 mm divergence, similar triangles give Earth’s radius at roughly 6,400 km — within a few percent of the true 6,371 km. A flat Earth predicts exactly zero divergence.
• 3It is not only bridges. The same accounting appears wherever structures get long: the 3.2 km Stanford linear accelerator is built straight through a curving Earth, London’s Crossrail tunnels were aligned for curvature, and LIGO’s 4 km arms (Entry 14) follow the chord, not the surface.

Sources: curvature in large structures (Verrazzano-Narrows Bridge) [94]. Kin to the horizon geometry of Entry 1 and LIGO’s 4 km arms (Entry 14). → Curvature data rows.

• 1Cavendish (1798). Lead spheres attract each other inside a sealed torsion balance, yielding G = 6.674×10⁻¹¹. Two inert weights pulling together in still air has no density or buoyancy explanation — it is gravitation, and it let us "weigh the Earth."
• 2Schiehallion (1774). A plumb line measurably deflects toward a mountain. Gravity points slightly sideways toward a large mass; pure "down = density" predicts exactly zero sideways pull.
• 3g is not uniform. It runs 9.780 m/s² at the equator to 9.832 at the poles and weakens with altitude; tides require a 1/r² pull from the Moon and Sun. A single upward acceleration would read identically everywhere and produce no tides.
• 4Newton → Einstein. Newton's inverse-square law (1687) predicts orbits; General Relativity (1915) refines it where they diverge — Mercury's 43″/century, starlight bent 1.75″ (1919 eclipse), redshift (Pound–Rebka 1959), gravitational waves (LIGO 2015).
• 5GPS proves it daily. Every receiver applies a ≈ +38 µs/day relativistic clock correction (SR −7, GR +45) or fixes drift ~10 km/day. Gravitational time dilation is a working engineering fact in your pocket.

Sources: Cavendish [16] · Schiehallion/Maskelyne [17] · tests of GR [18] · GPS relativity [19]. → Gravity data rows

• 1The buoyancy formula contains gravity. Archimedes: upward force = (fluid density) × (volume displaced) × g. That g is the strength of gravity. Density and buoyancy don't compete with gravity — they're a consequence of it.
• 2Why dense sinks. Gravity pulls all the fluid down, building a pressure that's higher at the bottom than the top. That pressure difference pushes lighter things up and lets heavier things settle. No gravity → no pressure gradient → no sorting by density.
• 3The clean test: free fall. In orbit or a drop tower, everything is weightless and buoyancy vanishes — a helium balloon doesn't rise, oil and water don't separate, a bubble just sits there. If "density" were a force of its own, it would still work in free fall. It doesn't, because it was gravity all along.
• 4It also can't explain the rest. Density says nothing about why g varies with latitude, why tides follow the Moon, why Cavendish's lab masses attract (Entry 9), or why GPS needs a relativistic clock fix. Gravity explains all of it; "density" explains only the order things settle in — once gravity is already doing the pulling.

Sources: Archimedes' principle [56] · gravity experiments [16],[18]. → Gravity data rows

• 1Vapour vs droplets. Most atmospheric water is invisible vapour — individual gas molecules fully mixed and buoyant in air. A visible cloud is condensed water: countless separate droplets ~10–20 µm across, not a connected sheet.
• 2Tiny means slow. A droplet's fall speed (Stokes' law) scales with the square of its radius: ~0.3 cm/s at 10 µm, ~1 cm/s at 20 µm. Gravity pulls fully; air drag — enormous relative to the droplet's mass — throttles the fall to a crawl.
• 3Air does the rest. Updrafts of even 0.1–1 m/s dwarf those fall speeds and carry droplets upward — and clouds form precisely in rising, cooling, buoyant air. The "millions of pounds" is about 0.5 g per cubic metre, ~0.04% of the air's own mass, spread through ~10⁹ m³.
• 4When it does fall. Let droplets collide and grow toward 1–5 mm and the same physics flips: fall speed climbs to 4–9 m/s and the water leaves the sky as rain. Nothing was ever suspended in defiance of gravity — it was a balance that tipped.

Same lesson as Entry 9: "it floats" is buoyancy and drag operating within gravity, not evidence against it — exactly like a balloon, a boat, or dust in a sunbeam.

Sources: cloud microphysics & Stokes' law [54]. → Water data rows

• 1"Gas fills a vacuum" assumes no gravity. That rule is for a sealed box where gravity doesn't matter over the size of the box. Over a whole planet, gravity pulls every layer of air downward; pressure is just the weight of the air above you. There's nothing to "rush into" because gravity keeps pulling it back.
• 2No edge — an exponential fade. Air pressure halves about every 5.5 km of altitude (scale height ~8.5 km). It never hits a wall; it just gets thinner and thinner until it's indistinguishable from space. The "edge of space" (Kármán line, 100 km) is a chosen convention, not a surface.
• 3It isn't perfectly sealed. The lightest gases — hydrogen and helium — slowly escape to space (Jeans escape). A true closed dome couldn't leak; a gravity-held atmosphere must, and ours does. That's the opposite of a sealed system.
• 4Every world does it. Mars, Titan, Venus, even the Sun's million-degree corona all hold gas against the surrounding vacuum by gravity alone — thick or thin, no container anywhere. Gravity bordering vacuum is the normal state of things, not a paradox needing a dome.

Sources: atmospheric structure, scale height & escape [58]. → Earth/atmosphere data rows

• 1Air is matter, and it has weight. A cubic metre of sea-level air masses about 1.2 kg. Stack kilometres of it and its own weight presses down: ~101 kPa at sea level — roughly 10 tonnes bearing on every square metre, balanced from all sides so you don't feel it. The ideal gas law, PV = nRT, ties that pressure to volume, temperature and amount of gas, and it's the workhorse behind weather, scuba tables and engines alike.
• 2The pressure gradient is gravity made visible. Pressure falls off exponentially with altitude — the barometric formula, with a scale height of ~8.5 km (Entry 12). Your ears pop in a lift or a climbing plane because the outside pressure drops as you rise; water boils cooler on a mountain; a sealed bag of chips puffs up at altitude. None of that happens unless gravity is pulling the gas downward into a deep, dense layer at the bottom.
• 3Heavy gases really do pool — sometimes lethally. Carbon dioxide, denser than air, collects in mine shafts, cellars and volcanic hollows; radon sinks into basements. In 1986 Lake Nyos belched a vast CO₂ cloud that flowed downhill like an invisible flood and suffocated about 1,700 people in the valleys below. Gravity sorts gases by density whenever mixing is slow — the open atmosphere stays blended only because thermal motion constantly re-stirs it.
• 4Gravity inside your chest. Stand up and gravity pulls blood and tugs lung tissue downward, so the base of each lung receives both more blood (perfusion) and more air (ventilation) than the apex — the classic "West zones" of respiratory physiology. It's why oxygenation shifts with posture and why the lower (dependent) lung is the better-perfused one. Every breath you take is a small, repeatable gravity experiment.

Sources: ideal gas law & kinetic theory [64] · barometric pressure & scale height [58] · pulmonary ventilation/perfusion gradient, West zones [65]. See also the atmosphere/dome entry (Entry 12) and buoyancy (Entry 10). → gas & pressure data rows

• 1It is a Michelson interferometer, grown enormous. LIGO splits a laser down two 4 km arms set at right angles, bounces it between mirrors (Fabry-Perot cavities) to fold the path out to ~1,120 km, and recombines the beams. A passing gravitational wave stretches one arm and squeezes the other by ~10⁻²¹ of their length; the recombined light shifts from dark to bright. Same principle as Entry 19 — just sensitive enough to feel spacetime itself flex.
• 2GW150914 — 14 September 2015. Two black holes of ~36 and ~29 solar masses spiralled together ~1.3 billion light-years away, merging into a ~62-solar-mass hole and radiating about 3 Suns' worth of mass-energy as gravitational waves in a fifth of a second. The chirp recorded at Livingston, Louisiana, and Hanford, Washington, matched each other and the relativistic prediction. The discovery won the 2017 Nobel Prize; hundreds of further mergers have been logged since.
• 3The 7-millisecond clue. The wave reached Livingston 7 ms before Hanford — precisely the time light needs to cross the 3,002 km between them. Two widely separated machines agreeing, with a delay set by the speed of light across a curved Earth, is exactly what a real, sky-sourced signal looks like and exactly what local noise or wishful thinking cannot fake.

Sources: LIGO & the first gravitational-wave detection [66]. Built on the interferometer of Entry 19; confirms the gravity of Entry 9. → LIGO data rows

• 1It's a difference, not just a pull. Gravity weakens with distance, so the ocean facing the Moon feels more pull than the solid Earth's centre, and the far ocean feels less — stretching the water into two bulges. The tidal (stretching) force falls off as 1/distance³, which is why the closer, smaller Moon out-tides the far larger Sun roughly 2:1. You need a real distance and a real diameter for this maths to work.
• 2The clock matches the sky. As Earth rotates under those two bulges, most places pass through both each day — two highs and two lows, arriving ~50 minutes later daily, tracking the Moon's own ~50-minute daily slippage. When Sun and Moon line up (new/full Moon) the bulges add for big "spring" tides; at right angles they partly cancel for "neap" tides. The pattern is the Moon's signature, written in water.
• 3Same force, bigger stage. The exact tidal mechanism — differential gravity across a body — is what whips material into accretion disks, tears comets apart, and powers the volcanoes of Jupiter's moon Io. It's gravity (Entry 9) seen through a magnifying glass, and it has no working substitute in any flat-Earth account.

Sources: tides & the tidal (differential gravity) force [73]. Built on the gravity of Entry 9. → tide & gravity data rows

• 1Steady motion is invisible to the senses. Your inner ear (the vestibular system) senses acceleration, not velocity. The only acceleration the spin adds is a steady centripetal ~0.034 m/s² at the equator — about 0.0035 g. The otolith organs don’t reliably register linear acceleration until roughly 0.06–0.1 m/s² (6–10 cm/s²), so the spin is below threshold — and, being constant, offers nothing to detect anyway.
• 2But we DO feel acceleration when it’s real — like an earthquake. The body is not numb to acceleration: a felt quake delivers abrupt, oscillating ground accelerations from a few percent of g up to ~0.25 g at Mercalli VIII (and more in severe shaking) — instantly noticeable. Turbulence, hard braking and a dropping elevator are felt for the same reason. The spin goes unfelt because it is smooth and sub-threshold, not because we can’t feel motion.
• 3The oceans stay put because gravity wins ~289:1. That outward 0.034 m/s² is about 1/289 of gravity’s 9.8 m/s². Its only real effect is that you weigh ~0.3% less at the equator than at the poles — measurable, and the opposite of flying off. To actually balance gravity at the equator the Earth would have to spin about 17× faster — a day of roughly 84 minutes.

Sources: vestibular acceleration thresholds [97]; earthquake ground acceleration [98]. The spin itself is measured directly in Entry 18 and Entry 21. → Rotation data rows.

• 1The hover trick fails on inertia. A helicopter lifting off keeps the eastward velocity it shared with the ground, so while hovering it continues east at the same rate as the land below and nothing slides past. There is no force that strips away that ~1,000 mph; stopping relative to the spinning Earth would take enormous, continuous thrust, not simply rising.
• 2The “1,000 mph runway” is a non-problem. The landing aircraft is also moving east at ~1,000 mph; relative to the runway only its airspeed (a couple of hundred mph) counts. Catching a tossed peanut on a moving train works for the identical reason — everything in the carriage shares the train’s velocity. Steady motion is undetectable from inside (Entry 16).
• 3The air comes along. Gravity holds the atmosphere down and friction drags it into rotation with the surface, so it co-rotates (the small leftover differences are ordinary winds). We do detect the spin — not by being flung off, but through subtle steady effects: deflected winds and shells (Coriolis, Entry 21), a turning pendulum (Foucault, Entry 18) and ring-laser gyros that sense it directly (Entry 20).

Sources: inertial frames & Galilean relativity [112]. Builds on why a steady spin is unfelt (Entry 16) and how it is actually detected (Entries 18, 20, 21). → Rotation data rows.

• 1Foucault's pendulum (1851). A long pendulum's swing plane slowly rotates as the Earth turns beneath it — first shown publicly by Léon Foucault under the Panthéon dome. No horizontal force turns the swing; the floor (Earth) is rotating.
• 2The latitude law. Precession = 15.04°/hr × sin(latitude): a full turn per sidereal day at the poles, none at the equator, in-between elsewhere (Paris ≈31.8 hr). That sin(latitude) signature is pure spherical geometry — a flat spinning disc wouldn't produce it. Try the model below.
• 3Coriolis. In a rotating frame, moving objects deflect — right in the Northern Hemisphere, left in the Southern (f = 2Ω sin φ). Hence cyclones spin counter-clockwise north of the equator and clockwise south of it; trade winds, ocean gyres, and long-range artillery all show it. The hemispheric reversal is a rotating-globe fingerprint.
• 4Not your sink. Coriolis is far too weak to decide which way a basin drains (that's set by basin shape and residual motion); it governs large, long-lived systems. The genuine, measured signatures are weather systems, the Eötvös effect (you weigh slightly less moving east), and ring-laser gyros sensing 15°/hr (Entry 22).

Interactive — precession by latitude

Drag the latitude. The swing plane rotates at 15.04°/hr × sin(latitude) — clockwise in the north, counter-clockwise in the south, frozen on the equator. (Animation sped up to be visible.)

Sources: Foucault pendulum [45] · Coriolis effect [46] · rotation rate [6]. → Rotation data rows

• 11881 → 1887: the hunt for the aether. Michelson built the first interferometer (Potsdam, 1881), then with Edward Morley (Cleveland, 1887) ran the definitive version on a stone slab floating in mercury, seeking the ~30 km/s "aether wind" of Earth's orbital motion. The result was null — no wind — the most famous null result in physics.
• 2What the null actually means. Not a stationary Earth: uniform motion has no detectable absolute effect because there is no aether and no preferred frame (special relativity, 1905). The geocentric/flat reading stops here; the physics didn't.
• 31925: detecting the spin. With Henry Gale and Fred Pearson, Michelson built a 1.9 km rectangular ring interferometer (612 × 339 m) on a field near Chicago. Rotation, unlike uniform motion, is absolute — and it produced a Sagnac fringe shift: predicted ≈0.236, observed 0.230 ± 0.005. Earth's rotation, measured with light. The prediction goes as A·Ω·sin(latitude), so the sphere is built in — a flat model gives a different, excluded number.
• 4Speed of light — where curvature entered. From the 1879 rotating-mirror work to the definitive 1926 Mount Wilson ↔ Lookout Mountain runs (~35 km), Michelson timed light over long baselines, getting c ≈ 299,796 km/s. To fix the distance, the U.S. Coast & Geodetic Survey laid the "Pasadena Base" (34.6 km) to ~1 part in 11 million and triangulated the path — a geodetic survey that necessarily reduces to the curved Earth. Curvature wasn't a nuisance; it was built into getting the answer.

Sources: Michelson-Morley 1887 [48] · Michelson-Gale-Pearson 1925 [49] · Michelson speed of light & Pasadena Base [50]. → Rotation & optics data rows

• 1Compton's water ring (1913). As a Wooster undergraduate, Arthur Compton filled a ring tube with water and oil droplets, let it settle, then flipped it 180°. Coriolis from Earth's spin sets the water drifting by a tiny, microscope-measurable amount — giving the rotation rate, the latitude, and the direction of true north, on a tabletop. (The third classical method, after the pendulum and the gyroscope.)
• 2The gyrocompass. Since ~1908 (Anschütz, Sperry), ships find true north not magnetically but by sensing the rotation axis itself; modern ring-laser and fiber-optic gyrocompasses do the same. Navies standardised on them precisely because they point to the geographic pole.
• 3Ring laser gyroscopes. The Wettzell "G" ring (Bavaria, 2002–) is a 4 m square laser Sagnac interferometer on a buried Zerodur block; it tracks Earth's rotation finely enough to see length-of-day changes below a millisecond, plus tides and polar motion (Nature Photonics, 2023). Every aircraft inertial system runs a smaller version (Entry 22).
• 4Entangled photons (2024). Philip Walther's group in Vienna ran a 2 km optical-fiber Sagnac interferometer with maximally entangled photon pairs and measured Earth's rotation on a two-photon quantum state — about 1000× better than prior quantum sensors and, in their words, "a century after the first observation of Earth's rotation with light" (Michelson-Gale, 1925). Classical light then, quantum light now — the same spinning planet.

Add Foucault's pendulum (Entry 18) and the Hafele-Keating flying-clock test (1971), and the spin has been independently confirmed by mechanics, fluids, optics, atomic clocks, and quantum entanglement — every one returning the same sidereal rate and the same sin(latitude) law.

Sources: Compton generator [51] · gyrocompass / inertial nav [21] · Wettzell G ring laser [52] · Hafele-Keating [53] · Walther/Vienna entanglement [47]. → Rotation data rows

• 1Storms pick a side. Hurricanes and low-pressure systems rotate counter-clockwise in the Northern Hemisphere and clockwise in the Southern — Buys-Ballot's law (1857). The deflection is zero at the equator and strongest at the poles, scaling with the sine of latitude, exactly as rotation predicts. A flat, non-rotating plane gives no preferred spin direction and no latitude dependence.
• 2Gunners and snipers pay for it. Long-range artillery fire-control and precision rifle solutions include an explicit Coriolis correction; ignore it past ~1,000 m and you miss measurably, and the sign of the correction flips between hemispheres. Militaries don't budget for a force that isn't there.
• 3Even a dropped stone drifts east. Because the top of a tall drop is moving east slightly faster than the bottom, falling objects land a touch to the east — measured in deep mineshafts by Ferdinand Reich in 1833 and many times since. It's a second, independent rotation signature alongside Foucault's pendulum (Entry 18) and the direct spin measurements of Entry 20; it's also why launch sites and aircraft inertial systems (Entry 22) account for it.

Sources: Coriolis effect, Buys-Ballot's law & deflection of falling bodies (Reich) [78]. Joins the rotation evidence of Entries 18 and 20. → rotation data rows

An aircraft holds a constant barometric altitude, which is a surface of constant geopotential — and that surface curves with the Earth. Lift always acts along the local vertical (perpendicular to the wings held level), and the local vertical itself rotates as you travel, at the rate v/R. At ~900 km/h that is about 0.0022°/s (~8°/hr) — a continuous, imperceptible nose-down that the autopilot and altimeter maintain without anyone touching anything. It never accumulates as felt pitch because the reference frame turns with it. On a flat plane, no such reorientation would ever be needed.

Interactive model — how an attitude indicator works

The instrument is a gyro-stabilized card erected to local vertical (gravity). The fixed orange aircraft symbol reads pitch (against the blue-sky / brown-ground split) and bank (against the top scale). Drag the sliders; then press play to fly around the globe and watch "level" silently re-define.

MEMS vs physical gyros — and what they reveal

• RLG/FOGOptical gyros (ring-laser, fiber-optic) use the Sagnac effect — no moving parts. Nav-grade bias is ~0.001–0.01°/hr, sensitive enough to measure Earth's 15.04°/hr rotation and gyrocompass to true north with no GPS. The airliner's IRS doesn't hide rotation — it depends on sensing it.
• MEMSMEMS gyros are vibrating-structure Coriolis sensors: tiny, cheap, ~1–100°/hr bias. Too noisy to resolve Earth rate, so they're GPS-aided. Found in light-aircraft AHRS, drones, and phones.
• SPINSpinning-mass gyros are the classic mechanical instrument behind legacy attitude indicators — erected to local vertical, drifting slowly and corrected continuously.

Sources: flight dynamics & attitude reference [23] · RLG/FOG & Earth-rate sensing [21] · MEMS gyros [22]. → Navigation data rows

• 1The curve rate is imperceptible. Speed ÷ Earth radius gives the rate the local vertical turns: ~250 m/s ÷ 6,371 km ≈ 0.0022°/s. No panel shows that as a “dip,” and no one feels a steady 8°-per-hour rotation. Cruise pitch attitude is set by angle of attack for lift, not by curvature, which the autopilot trims out continuously.
• 2Pressure altitude follows the geoid. Air pressure falls with height, so a chosen pressure level is a smooth shell wrapped around the round Earth — the curved “level” surface of Entry 4. Flying a constant pressure altitude keeps the aircraft on that shell, quietly tracking the curve with no dramatic nose-down.
• 3Why 29.92 above 18,000 ft. Below the US transition altitude pilots set local sea-level pressure so the altimeter reads true height for terrain clearance; at and above 18,000 ft everyone switches to the standard 29.92 inHg (1013.25 hPa) and flies “flight levels,” a single shared datum that guarantees vertical separation regardless of local weather. The same avionics run on a rotating WGS-84 ellipsoid, and their ring-laser gyros sense Earth’s spin (Entry 20).

Sources: flight levels & the 29.92 inHg standard setting [111]. Extends the attitude-indicator model of Entry 22 and the curved “level” surface of Entry 4. → Navigation data rows.

• 1West vs East. The atmosphere rotates with the Earth, so spin gives no westward advantage. The ~1 h JFK↔LHR difference is the jet stream — a west-to-east mid-latitude wind (~110 kt). Eastbound rides it; westbound fights it.
• 2Great circles. The shortest path on a sphere. On a flat Mercator map they appear as curves arcing toward the pole — which is exactly how real flight tracks look. The "weird" curve is a globe signature.
• 3Southern routes settle it. Santiago–Sydney, Johannesburg–Perth and Santiago–Johannesburg are short, direct great circles on a globe. On the flat-Earth north-pole azimuthal map those same city pairs become enormous detours — yet the flights run on schedule at sensible durations.
• 4Antarctica is not a wall. It's overflown by research and military aircraft and occasionally airliners — Qantas QF28 reached ≈74° S in 2023 (787-9, ETOPS-330); the A350 is ETOPS-370. Routine routing is blocked by ETOPS diversion-airport rules, extreme weather, and the fact that few city pairs' great circles cross it — not by an edge of the world. The Arctic, with diversion fields (Anchorage, Iqaluit, Keflavík, Svalbard), is overflown daily.

Sources: jet stream [24] · great-circle navigation [25] · Arctic/Antarctic overflight & ETOPS [26]. → Navigation data rows

• 1Going around closes the loop. An east–west circumnavigation totals ~40,000 km near the equator and less along higher-latitude parallels — a fixed, sphere-specific budget of distance — and you arrive back where you began, which an infinite plane or an ice-rimmed disc cannot reproduce.
• 2Great circles look “curved” on flat maps. The shortest path on a sphere bows poleward when drawn on a flat projection. That is why a Hong Kong–Los Angeles flight passes near Alaska, and why a diversion there is the closest airport, not a detour. On a globe the routes are straight; the “detour” is an artifact of the map.
• 3The book backfires. Its cases sit on northern great circles, which the pole-centred azimuthal (“Gleason”) map happens to render nearly straight — so it cherry-picks the hemisphere where that map is least wrong. The same map balloons the south: Sydney–Santiago and Perth–Johannesburg become far longer than their real, on-schedule flight times allow (Entry 26). One map cannot be right in the north and impossible in the south.

Sources: circumnavigation & great-circle routing [108]. Connects to the route geometry of Entry 24 and the polar flights of Entry 26. → Navigation data rows.

• 1Both poles, one loop — ratified. In July 2019 the “One More Orbit” Gulfstream G650ER circled the globe over both poles in 46 hours 40 minutes (~40,000 km), GPS-logged and certified by the FAI and Guinness. It was not the first: a Boeing 707 (“Pole Cat”) did it in 1965, and a 1968 flight crossed both poles and landed at McMurdo — the first aircraft to touch all seven continents.
• 2Antarctica is flown over and into. Qantas has run sold-out sightseeing overflights from Australia since 1977 (Boeing 787, ~13½ hours, with hours spent above the continent). Charter airliners fly Punta Arenas–King George Island in ~2 hours, and ski-equipped aircraft serve interior blue-ice runways at Union Glacier and Wolf’s Fang, with flights onward to the South Pole.
• 3And toured by the thousands. Over 100,000 tourists a year now visit, the great majority by ship to the Antarctic Peninsula from Ushuaia or Punta Arenas; yachts have circumnavigated the continent since the 1970s. None of it is secret or guarded — it is booked online and regulated under the Antarctic Treaty (Entry 60).

Sources: the One More Orbit polar circumnavigation [109]; Antarctic overflights & tourism [110]. Pairs with the circumnavigation geometry of Entry 25 and the Antarctica claims of Entry 60. → Navigation data rows.

• 1No monopoles exist. Maxwell's ∇·B = 0 says magnetic field lines have no isolated source or sink. Every magnet — and the Earth — has two poles. Cut a bar magnet in half and you get two new dipoles, never a lone pole.
• 2Three kinds of poles, each N and S. Geographic (the rotation axis, fixed at 90°); magnetic dip (field vertical — the north one is now drifting from Canada toward Siberia per WMM2025; the south sits off Antarctica); and geomagnetic (the best-fit dipole axis). They don't coincide — the gap between magnetic and true north is the declination.
• 3Inclination settles it. Magnetic dip runs from +90° (straight down) at the north dip pole to −90° (straight up) at the south dip pole. A single central pole could pull field lines only one way; the observed reversal of the vertical field between hemispheres requires two poles.
• 4The geodynamo. The dipole is generated by convecting liquid iron in the outer core; it drifts continuously and reverses irregularly (183 times in 83 Myr). A static disc with one central pole explains none of this.

Sources: monopoles / Maxwell [32] · geomagnetism & WMM2025 [33]. → Magnetism data rows

A total solar eclipse is total only along a narrow track — about 150–290 km wide for TSE 2026 — even though the Sun and Moon are enormous. That narrow band exists only because the Sun is ~400× larger and ~389× farther than the Moon (see the Relationships rows), so the Moon's shadow cone tapers almost exactly to a point at Earth's distance. The umbra's tip grazes the surface and sweeps a thin line. A small, nearby Sun — which flat-Earth models require, typically ~50 km across and ~5,000 km up — cannot cast a sharp 150 km umbra at all; that geometry produces a vast, soft penumbra or no totality whatsoever.

Why TSE 2026 cannot occur on a flat Earth:

• 1Retrograde shadow. The first half of the path runs east-to-west from Arctic Russia toward Greenland. NASA's own description: near the pole, at the Moon's descending node, Earth's rotation cancels the umbra's normal eastward motion. There is no arrangement of a circling flat-Earth Sun and Moon that drives a shadow backward across the map and then curves it south.
• 2Sunrise/sunset loops. The path ends in closed loops where the umbra grazes the day/night terminator. A terminator is the edge of a sphere's lit hemisphere — flat-Earth models rely on a "spotlight" Sun that produces no sharp terminator and no such loops.
• 3Simultaneity across a continent-spanning arc. The same eclipse is at local mid-day in Arctic Russia yet only a few degrees above the horizon near sunset in Spain, with the Sun never above ~26° anywhere on the track. That requires near-parallel rays from a distant Sun onto a curved surface. A nearby Sun would appear at wildly different, inconsistent altitudes and the shadow could never form one continuous narrow path.
• 4Predicted to the second, decades out. The path and timings were computed from heliocentric spherical ephemerides (VSOP87 / ELP2000) and Besselian elements — to ~1 s and ~1 km. Eclipses recur on the Saros cycle (18 yr 11 d 8 h), a globe-geometry phenomenon. No flat-Earth model has ever predicted an eclipse's track or time.
• 5Companion proof — lunar eclipses. Earth's shadow on the Moon is always circular, from every orientation across the year. Only a sphere casts a round shadow from every angle; a disc would project an ellipse or a line at most geometries. Aristotle made this argument ~350 BCE.

Result. Every measurable feature of TSE 2026 — width, direction, altitude profile, timing, recurrence — follows from a Sun ~400× larger and far away, a Moon casting a needle-tip umbra, and a rotating sphere. None of it is reproducible on a flat plane. → Eclipse data rows

Limb-resolved prediction — re-entrant totality in Greenland

Standard eclipse maps treat the Moon as a smooth ball. Feed in its real topography — the mountains and valleys on the limb, mapped from lunar orbit — and the umbra's outline stops being a clean ellipse: it becomes a jagged polygon whose edges bow slightly inward and meet at cusps. Recomputing the true-limb path limits for TSE 2026, eclipse computer John Irwin (Besselian Elements) found a tiny zone — about 150 km south-southwest of Station North on the Princess Ingeborg Peninsula, Kronprins Christian Land, northern Greenland, right at the path edge — where the Sun's limb pivots through the bottom of a single narrow, deep lunar valley. There, totality is predicted to last ~9 s but be interrupted for ~2 s by a near-imperceptible return to partiality: two second-contacts and two third-contacts. That is re-entrant totality — in the smooth-Moon model it cannot happen at all.

Sources: NASA Science TSE-2026 [12] · NASA GSFC / Espenak Besselian elements [13] · NASA SVS path visualization [14] · perigee & path notes [15] · re-entrant totality, J. Irwin / Besselian Elements [27].

• 1Always an arc, every single time. Across hundreds of recorded eclipses, the shadow's curved bite on the Moon is consistent with a circle of the same radius each time. A flat disc tilted to the Sun would usually cast an oval or a thin line; a sphere is the only solid whose silhouette is a circle from all directions.
• 2The geometry is fixed and predictable. The shadow's curvature implies an Earth roughly 3.7× the Moon's diameter — consistent with timing how long the Moon takes to cross the shadow. The same Sun–Earth–Moon geometry that produces this also predicts eclipse dates and the lunar-eclipse "blood Moon" colour from refracted sunlight (see eclipses, Entry 28).
• 3No 'shadow object' needed. The eclipsing body is always opposite the Sun, moves at exactly the Moon's orbital rate, and never transits at any other time — because it is the Earth's own shadow. Invoking an unseen disc-shaped intruder that only ever appears anti-solar is an unfalsifiable patch, not an explanation (Entry 1 covers the surface curvature this complements).

Sources: shape of Earth's shadow during lunar eclipses (Aristotle's argument) [74]. Pairs with the eclipse geometry of Entry 28. → eclipse data rows

• 1Refraction lifts both bodies ~0.5–0.6°. The same bending that lets you see the Sun for a few extra minutes after it has geometrically set lifts the eclipsed Moon too. Because the lift (~34′) exceeds each body’s ~16′ radius, both can sit fully above their opposite horizons briefly, even though their true positions are below.
• 2It is confined to sunrise/sunset, lasts minutes, and needs an open horizon. A selenelion can be seen only just after sunrise or just before sunset, near opposite points of the sky, often best from a high ridge — precisely the narrow conditions a refracting spherical atmosphere predicts. A flat model with a nearby Sun and Moon has no clean reason for the timing or the brevity.
• 3The shadow still behaves. Throughout, Earth’s shadow crosses the Moon from the geometrically correct side and the eclipse contacts match prediction to the minute (Entry 29). Pliny the Elder noted the effect about 2,000 years ago — it is old, quantified and well understood, not an anomaly.

Sources: selenelion & horizon refraction [103]. Builds on the round shadow of Entry 29 and the refraction that shapes the horizon (Entry 4). → Eclipse data rows.

• 1We predict the unprecedented. In 1846 Le Verrier and Adams computed an unseen planet’s position from tiny wobbles in Uranus’s orbit, and Neptune was found within ~1° of the prediction; Halley used gravity to predict a comet’s return decades after his own death. Neither was “we’ve seen this before” — they were forecasts of events no one had recorded.
• 2Cycles can’t place a shadow. Knowing an eclipse recurs every ~18 years does not tell you it will go total over a named town at a stated minute — yet that is exactly what is published years ahead and confirmed by millions (Entries 28–29). Pinpointing the path needs the real three-dimensional geometry and motion of Sun, Earth and Moon.
• 3The same model flies spacecraft. Probes are launched to arrive at a precise point years later, using gravity assists timed to the second; an asteroid’s occultation of a star is predicted to a particular roadside for a particular few seconds. Cycle-counting could never thread those needles — a working physical model of a moving system can (Entry 35).

Sources: the gravitational prediction of Neptune [113]. Builds on the eclipse geometry of Entries 28–29 and the modelled solar system of Entry 35. → Eclipse data rows.

• 1Dark, not bright. Geometric albedo ~0.12, Bond ~0.11 — comparable to worn asphalt; the maria are darker (~0.07) than the highlands (~0.11–0.18). The Moon looks bright only because it is large and close. A body making its own light would not wax and wane in step with the Sun.
• 2Retro-reflective regolith. The powdery, shadow-hiding surface scatters light preferentially straight back toward the Sun (the opposition surge), so a full Moon is far more than twice a half Moon and the disc is lit evenly edge to edge. A simple matte (Lambertian) sphere would show limb-darkening; the regolith cancels it — which is exactly why a real sphere can look like a flat disc.
• 3Phase-locked to the Sun. The illuminated fraction always faces the Sun, and the Moon goes dark in eclipse. Both are impossible for an independent light source and automatic for a sunlit ball.

Sources: lunar albedo & regolith reflectance [85]. Pairs with the retroreflector ranging of Entry 57. → Moon data rows.

• 1Earth as a mirror. From the Moon a “full Earth” is about 50× brighter than a full Moon is to us — easily enough to faintly light the lunar night. The glow is doubly reflected (Sun → Earth → Moon → eye), which is why it is so dim.
• 2The phases interlock. Earthshine is brightest near new Moon (a thin crescent here means a full Earth there) and fades as the Moon waxes (Earth wanes as seen from the Moon). That anti-correlation is exact two-ball-in-sunlight geometry, not self-illumination.
• 3It measures Earth’s albedo. Astronomers monitor earthshine to track Earth’s reflectivity and cloud cover over time — a measurement that only makes sense if Earth is a sunlit body reflecting onto the Moon.

Sources: earthshine / Da Vinci glow [87]. Builds on the reflective regolith of Entry 32. → Moon data rows.

• 1More than half. Optical libration in longitude (from the eccentric orbit) and latitude (from axial tilt), plus diurnal libration (Earth’s own radius shifting our viewpoint), together expose ~59% of the surface — features near the limb swing into and out of view month to month.
• 2Constant spin, varying orbital speed. The Moon turns once per orbit at a steady rate, but Kepler’s second law speeds it near perigee, so its face appears to lead and lag. That apparent rocking is something only a rotating body on an ellipse can produce.
• 3Measured to the metre. Lunar laser ranging tracks the Moon’s physical libration (real nodding from its slightly non-spherical mass), which probes the lunar interior — see Entry 57.

Sources: lunar libration [88]. Ties to retroreflector ranging (Entry 57). → Moon data rows.

• 1What was actually proven. Two bodies have an exact Kepler solution. Poincaré (1890) showed three-or-more bodies are non-integrable — no general formula — and discovered deterministic chaos in the process. Neither result says the motion is unknowable.
• 2We compute it numerically. n-body integration (JPL's DE440 ephemeris) propagates every major body forward step by step, accurate enough to land probes on comets and to predict TSE 2026's path to ~1 s and ~1 km.
• 3Exact special solutions exist anyway. Euler's collinear (1767) and Lagrange's equilateral (1772) configurations give the five Lagrange points — JWST orbits Sun–Earth L2 right now. The figure-eight three-body orbit was found in 2000.
• 4Chaos ≠ random. Same inputs give same outputs; chaos only limits prediction over millions of years — far beyond the decades and centuries that eclipse forecasting and spaceflight require.

Sources: Poincaré, Lagrange, JPL ephemerides [20]. → Gravity/orbital data rows

• 1Two opposite centers. Northern star trails circle Polaris counter-clockwise; southern trails circle σ Octantis clockwise. Below the equator the whole sky turns about a southern point. One plane under one dome cannot produce two opposite rotation centers on opposite sides of the world.
• 2The southern pole star is real. σ Octantis (Polaris Australis), ~mag 5.4 — faint, but it marks a physical south celestial pole, not a void.
• 3The pole star changes. Earth's axis precesses on a ~25,772-yr cycle. Polaris is near the pole now (closest ~2100 AD); Thuban held the role ~2700 BC; Vega will ~13,700 AD. The southern pole star cycles too. A wobbling spinning oblate planet does this; a flat disc has no analogue.
• 4Measurable today. Precession (~50.3″/yr) shifts the equinoxes and the pole steadily — exactly the motion of a torqued spinning top, observed and tabulated for centuries.

Sources: pole stars & axial precession [34]. → Pole-star data rows

• 1It circles the pole nightly. Offset ~0.7° from the exact pole, Polaris traces a small arc each night; stars nearer the true pole trace smaller circles and those farther off trace wider ones. A camera pointed north for an hour records Polaris as a short curved streak, not a fixed dot.
• 2The pole star changes over time. Precession swings the axis around a ~25,800-year circle: Thuban (in Draco) was the pole star around 3000 BCE, when the pyramids were built; Vega will take the role around 13,700 CE. Polaris is merely our current, temporary marker, closest to the pole about 2100 CE.
• 3Near-fixed is what a spin axis predicts. A star almost on the rotation axis barely moves while stars farther off sweep wide circles — and the Southern Hemisphere, whose sky turns about a different point, has no bright pole star at all (Entry 36). A flat Earth under a dome gives no reason for a star’s altitude to equal your latitude, or for a second southern centre of rotation.

Sources: Polaris’s offset & axial precession [114]. Extends the two-centres-of-rotation argument of Entry 36 and the star trails of Entry 43. → Sky data rows.

• 1Parallax — the nearby ones lean. As Earth swings across its ~300-million-km orbit, close stars appear to shift against the far background, just as a finger shifts against the wall when you blink each eye. The shift is minuscule (61 Cygni's 0.314″ is like a 2 cm move seen from 12 km away), which is exactly why it took until 1838 to detect — and why its long absence was once used to argue against a moving Earth. The closer the star, the bigger the shift; that distance-dependence is the fingerprint.
• 2Aberration — all of them tilt. Bradley found every star drifts in a ~20.5″ yearly ellipse regardless of distance — not parallax, but the result of Earth's velocity adding to light's, like rain slanting on a moving car's windscreen. Airy's 1871 water-filled telescope confirmed it: slowing the light inside the tube didn't change the angle, ruling out aether-drag and matching a genuinely moving observer.
• 3Two independent witnesses to orbital motion. Parallax depends on distance; aberration doesn't — so they can't both be coincidences of some local effect. Together they nailed Earth's orbital motion long before spaceflight, and they sit naturally beside the Michelson–Morley result (Entry 19), which probed that same 30 km/s motion.

Sources: stellar parallax (Bessel, 1838) [76] · aberration of light (Bradley 1727; Airy 1871) [77]. Companion to Michelson–Morley (Entry 19). → stellar-motion data rows

• 1What it is. Mark the Sun's position at, say, noon every week for a year. The dots form a tall, skewed figure-8 in the sky — the analemma.
• 2Two causes, both global. The tall (up-down) extent comes from Earth's 23.4° axial tilt — the Sun rides high in summer, low in winter. The side-to-side width comes from the equation of time: Earth's orbit is an ellipse, so the Sun runs up to ~16 minutes ahead of or behind clock time through the year.
• 3It's planet-specific. Mars traces a teardrop, not a figure-8, because its tilt and orbit differ — and rovers have photographed it. A flat plane under a circling local Sun produces no such curve; the analemma only falls out of a tilted globe in an elliptical orbit.
• 4It's also why sundials disagree with clocks. The east-west swing of the analemma is the equation of time engraved on every accurate sundial.

Sources: analemma & equation of time [55] · Sun facts [3]. → Pole-stars/sky data rows

• 1Constant size = enormous distance. Measured with a safe solar filter, the Sun is ~0.5° wide all day (varying only ~3% over the year from Earth's elliptical orbit). On a local-spotlight disc, the Sun's distance would change by thousands of kilometres between noon and sunset, shrinking it to a fraction of its size. It doesn't shrink — because it's so far away that crossing the sky barely changes the distance.
• 2Sunsets go bottom-first. The Sun (and Moon, and ships) vanish from the bottom up as they cross a crisp horizon line — the hallmark of going behind a curve. A spotlight receding on a flat plane would shrink toward a point near eye level and never be occluded from below. Perspective never makes a whole object disappear edge-first while keeping its size.
• 3And it would never fully set. On a flat Earth with a circling Sun, the Sun would always be above the plane somewhere and merely get small — true darkness and a clean horizon-set are impossible. Combined with the day/night terminator (Entry 44), the yearly analemma (Entry 39), and CME timing of the real distance (Entry 45), the local Sun has nowhere left to hide.

Sources: Sun's angular diameter & apparent motion [75]. See the terminator (Entry 44), analemma (Entry 39), pole stars (Entry 36) and CME distance (Entry 45). → Sun data rows

• 1What the law says. Energy from a source spreads over a sphere whose area grows as r², so the intensity you receive falls as 1/r². The same geometry governs gravity (Newton’s 1/r² law, Entry 9), which is why tides — driven by the difference in pull across the Earth — fall off faster still, as 1/r³ (Entry 15).
• 2A local Sun fails two tests at once. Apparent size scales as 1/r and received light as 1/r². A 5,000-km-high sun would noticeably shrink and dim between overhead and the horizon; the real Sun holds a steady ~0.5° and ~1,361 W/m² (top of atmosphere) all day. Sunlight on Mauna Kea’s summit and at sea level differs immeasurably, because a few kilometres is nothing against 150 million (Entry 40).
• 3Done right — not the cartoon version. The careless flat-Earth form (“brightness → infinity as distance → 0, so the Moon would blind astronauts”) misuses the law: the surface brightness of an extended source is distance-independent — only its angular size, and hence the total light it delivers, changes. Applied correctly, the law still kills the local Sun and pins it near one astronomical unit.

Sources: inverse-square law & solar irradiance (~1,361 W/m² at 1 AU) [106]; the inverse-square law of light [107]. Connects to the constant solar size of Entry 40 and the 1/r² gravity of Entry 9. → Sun data rows.

• 1Spreading, not dying. The inverse-square law thins light over distance; it never sets it to zero. Collect more of it — a wider aperture, a longer exposure — and the faint signal builds up. Ordinary long camera exposures already reveal stars far too dim for the eye; observatories simply do this on a grand scale.
• 2We see across the universe. Sunlight reaches us in ~8 minutes, Andromeda’s in ~2.5 million years, and deep-field images (Hubble, JWST) capture galaxies whose light left them over 13 billion years ago. Cosmic expansion stretches that light to longer wavelengths (redshift) but does not extinguish it.
• 3What actually dims starlight. Real losses come from intervening matter — interstellar dust and gas absorbing or scattering light — not from photons exhausting themselves; astronomers map that dust and see through it in the infrared. A “local” Sun and stars are independently ruled out by parallax (Entry 38) and the inverse-square test (Entry 41).

Sources: deep-field imaging & cosmic distances [115]; inverse-square dimming [106]. Pairs with stellar parallax (Entry 38) and the local-Sun refutation (Entry 41). → Sun data rows.

• 1North. A camera left open toward Polaris records concentric arcs turning counter-clockwise about the north celestial pole.
• 2South. The same exposure below the equator shows arcs about σ Octantis turning clockwise — a second, independent center of rotation.
• 3Equator. Aim at the celestial equator and the trails are nearly straight lines rising vertically, then setting. A single overhead center on a flat disc cannot produce straight equatorial trails and two opposite circular centers.
• 4The arc length tells the period. In a known exposure, every star sweeps the same angle — 15° per hour — because the whole sky reflects one rotation: the Earth's, once per sidereal day.

Sources: celestial poles & rotation [34] · rotation rate [6]. → Pole-stars data rows

• 1Half-lit, give or take. The terminator — the day/night line — is a great circle that cuts Earth into a lit half and a dark half. A little over half, about 50.3–50.5%, is sunlit at any moment, because the Sun has a finite width (~0.5°) and the air bends its light ~0.6° at the horizon. It is never 70%, never 90% of the surface.
• 2Why the map says otherwise. Three real effects stack up. (a) Projection: equirectangular and Mercator maps blow up high latitudes, so the lit cap looks enormous. (b) Land asymmetry: the land hemisphere is centred opposite the Pacific, so when it faces the Sun almost every continent is in daylight at once — "most of the world" by land or population, still half the surface. (c) Solstice tilt: near June or December the terminator tilts 23.4°, one pole drowned in midnight sun and the other in polar night, so the lit region sweeps pole-to-pole down one side of the map.
• 3The antipode test (the clean one). For any point, the Sun's altitude at its antipode is exactly minus its altitude at the point. The instant the Sun stands 40° above your horizon, it is 40° below the horizon at your antipode — day here is night there, every day of the year. The only shared instant is when both points sit exactly on the horizon: a single simultaneous sunrise/sunset. A flat disc lit by a circling spotlight cannot reproduce this clean ±symmetry. Run it below.
• 4"Opposite sides" usually aren't. People picture America and Asia as the two ends of the ball, but New York's antipode is empty Indian Ocean southwest of Australia; Beijing's is Argentina; London's is the aptly named Antipodes Islands near New Zealand. Only ~15% of land has land at its antipode (≈4.4% of the surface) — the rest is sea. So two distant cities both in daylight is ordinary, not paradoxical: they simply are not antipodal.

Sources: terminator & illuminated fraction [62] · antipodes & the land/water split [63]. See also eclipses (Entry 28) and the analemma (Entry 43). → day/night data rows

• 1See it leave, time its arrival. Coronagraphs — chiefly LASCO aboard the SOHO spacecraft, parked at the L1 point ~1.5 million km sunward of Earth — record CMEs erupting and clock their speed, often 300–3,000 km/s. Forecasters then predict the strike on Earth's magnetic field to within hours, and the geomagnetic storm arrives on schedule. The fastest ever recorded, the 1859 Carrington event, crossed in about 17 hours; most take 1–3 days.
• 2The arithmetic only closes at 150 million km. A 1,000 km/s CME taking ~42 hours covers ~150 million km — one astronomical unit. Run the same sum for a Sun 5,000 km overhead and the blast would arrive in about five seconds. Space-weather agencies stake real power-grid and satellite decisions on the multi-day number, and it works. (The distance is also pinned independently by radar ranging of the inner planets and by parallax.)
• 3And we have flown a probe there. NASA's Parker Solar Probe has repeatedly dived to 3.8 million miles (6.1 million km) of the Sun's surface at 430,000 mph — the fastest object humans have built — sending its data home through the Deep Space Network (Entry 58). You cannot fly a spacecraft for years to a destination that is only a few thousand kilometres up.

Sources: coronal mass ejections & CME transit times (SOHO/space weather) [71] · Parker Solar Probe [70]. See also eclipses (Entry 28), the day/night terminator (Entry 44) and the Deep Space Network (Entry 58). → Sun-distance data rows

Frequency decides how a signal gets past the horizon — and every mechanism that beats the horizon is itself evidence of one:

• LF/MFGround wave. Long-wave and AM diffract along the conductive surface, hugging the curve for hundreds of km with no line of sight.
• HFSkywave. 3–30 MHz refracts off the ionosphere's F layer and returns to ground far over the horizon; multi-hop gives worldwide reach (ham DX). The D layer absorbs HF by day, so skywave favours night.
• VHF/UHFLine-of-sight, extended. Normally horizon-limited, but tropospheric ducting (a temperature-inversion waveguide — the 4/3-earth refraction gradient taken to an extreme) and Sporadic E (patchy E-layer ionization) carry VHF hundreds to ~2,000 km past the horizon.

Marconi & the invention of the ionosphere

In December 1901 Marconi sent a signal from Poldhu, Cornwall to Signal Hill, Newfoundland — about 3,500 km. Physicists expected it to fail: radio was thought to travel straight, and over that distance the receiver sits more than 100 km below the line of sight, hidden by the bulge of the Atlantic. It worked anyway — which is exactly why Kennelly and Heaviside (1902) independently proposed a reflecting layer high in the atmosphere, later named the ionosphere and confirmed by Appleton in the 1920s. The ionosphere was hypothesized because Earth's curvature made straight-line transatlantic radio impossible. The flat-Earth reading inverts the actual history.

LoRa / LoRaWAN — long range, and what sets its limit

LoRa uses chirp spread-spectrum (CSS) modulation in sub-GHz ISM bands. The processing gain buys enormous receiver sensitivity (≈ −137 dBm) at low data rates, so a 25-mW node reaches remarkably far. The headline records — ≈832 km on 25 mW, and ≈1,336 km reported — were all set by lofting the node on a high-altitude balloon (~38 km). Ground-to-ground links top out near ~212 km from mountaintops and towers. The gap is the radio horizon: d(km) ≈ 4.12(√hₜ + √hᵣ), h in metres. A balloon at 38 km has a horizon near 800 km on its own — which is the record. LoRa's distance feats are a direct measurement of the curve: to beat the horizon you have to climb above it.

The Fresnel zone

Even within line of sight, a link needs a clear elliptical first Fresnel zone around the path — r ≈ 17.3·√(d₁d₂/(f·d)) metres — kept roughly 60% unobstructed or the signal fades. On long paths the Earth's own bulge rises into that zone, so microwave-tower heights are computed from earth curvature (the 4/3 model) plus Fresnel clearance. Designing around the curve isn't controversial in RF engineering; it's day-one path-budget math.

Sources: HF ground/skywave [35] · ionosphere [36] · tropospheric ducting & Sporadic E [37] · Marconi & Kennelly–Heaviside [38] · LoRa/LoRaWAN records [39] · Fresnel zone [40] · radio horizon (4/3) [9]. Pairs with over-the-horizon radar (Entry 47) & microwave links (Entry 48). → Radio data rows

• 1Conventional radar stops at the horizon. Line-of-sight microwave radar is blocked by Earth’s bulge — a low-flying target drops below the radar horizon within tens of kilometres. There is nothing to “beat” on a flat plane; the very existence of a radar horizon is a curvature measurement.
• 2OTH-B bounces over the bulge. Skywave over-the-horizon radar refracts HF signals (3–30 MHz) off the ionosphere (~100–300 km up) and back down 1,000–3,000 km away — up to ~6,000 km with a double hop. The beam is aimed just 2–4° above the horizon and needs antenna arrays 2–3 km long. The whole design is geometry computed for a spherical Earth under a curved ionospheric shell.
• 3Real, deployed, and current. Australia’s JORN, the US Navy’s ROTHR, the Soviet “Duga” and Russia’s Container all work this way; in 2025 Canada agreed to buy JORN technology for Arctic coverage. None of it would be necessary — or even make sense — over a flat plane with an unobstructed line of sight.

Sources: over-the-horizon radar & the radar horizon [93]. Pairs with ionospheric skywave (Entry 46) and moonbounce (Entry 50). → Radio data rows.

• 1The Earth bulges into the path. On a 30 km hop the surface rises ~13 m at the midpoint (bulge ≈ d₁·d₂ / 12.74K, in metres); on longer hops it is tens of metres. Towers are sized specifically to clear that bulge and keep 60% of the first Fresnel zone open. There is nothing to clear on a flat plane.
• 2The 4/3-Earth-radius rule. Path profiles are drawn on an “effective Earth” 4/3 the true radius — the standard correction for how the atmosphere bends the beam gently downward (the K-factor). Engineers literally plot the link against a curved Earth before a single tower goes up.
• 3The dishes point slightly downhill. Over a 50 km hop between equal-height towers, each antenna is aimed about 0.2° below its own local horizontal (≈ half the path’s central angle, d/2R), because the far tower has dropped below it around the curve. Aim both dead level and the link fails.

Sources: line-of-sight propagation, earth bulge & the 4/3-Earth-radius rule [95]. Sits beside the radio horizon of Entry 46 and over-the-horizon radar (Entry 47). → Radio data rows.

• 1Light speed is the speed limit. Light in glass fibre moves at ~200,000 km/s (the fibre's refractive index ~1.47 slows it from 300,000). So 20,000 km one way takes at least ~100 ms, ~200 ms round trip — before any routing overhead. You cannot ping faster than physics allows, which turns latency into a ruler.
• 2The numbers match the globe, not the map. Measured RTTs between far-apart cities track great-circle distances on a sphere. On the common flat "azimuthal" map, places like Sydney and Santiago are drawn vastly farther apart than the globe puts them — yet their real latency matches the short globe distance. Submarine-cable maps (which follow great circles) and CDN routing are built entirely on the spherical figures.
• 3Run it yourself. Ping servers on several continents, halve the RTT, multiply by ~200,000 km/s, and you recover continent-scale distances that only close on a globe. It's the modern, self-serve cousin of the satellites of Entry 53 and the Deep Space Network's light-delays (Entry 58), needing nothing but a terminal (and it underlies the radio links of Entry 46).

Sources: speed of light in optical fibre & great-circle network latency [79]. Kin to satellites (Entry 53) and DSN light-time (Entry 58). → latency data rows

• 1The echo time gives the distance. Light (and radio) covers ~300,000 km/s, so a ~2.56-second round trip means ~768,000 km total — ~384,000 km to the Moon. The delay varies slightly through the month as the Moon's distance changes (perigee to apogee), and the variation matches the predicted orbit. That's distance measured directly by the clock, not assumed.
• 2It's old, open, and repeatable. The US Army first bounced radar off the Moon in 1946 (Project Diana); hams have done it since the 1950s, and today it's a recognised operating mode for long-distance contacts. The technique, antenna sizes and link budgets are all public — you can hear your own voice return after its lunar detour.
• 3A near Moon can't carry the signal. The path loss, Doppler shift and 2.5-second delay all only add up for a target ~384,000 km away. The same distance is independently confirmed by laser ranging off the Apollo retroreflectors. It's the radio cousin of the satellite and deep-space links (Entries 53 and 46) — just aimed at the Moon.

Sources: Earth–Moon–Earth (EME / moonbounce) communication [80]. Related to satellites (Entry 53) and radio propagation (Entry 46). → Moon-distance data rows

• 1Long-path: the signal arrives from "behind." Point the beam 180° away from a station and you can still work it — stronger, sometimes, than the direct route — because the signal circled the planet the long way via repeated ionospheric hops. The arrival bearing and the ~extra delay correspond to going around a ~40,000 km sphere. On a flat map there is no "other way around."
• 2Gray-line: riding the terminator. Along the moving sunrise/sunset line, the ionosphere's absorbing D-layer fades while the reflecting layers persist, opening a low-loss duct. Operators schedule contacts for the minutes the gray line links their two locations — a propagation window that tracks the day/night terminator sweeping around a rotating globe (Entry 44).
• 3Curvature is baked into the hobby. Great-circle beam headings, sunrise/sunset tables and ionospheric skip (Entry 46) are standard tools precisely because the Earth is a rotating sphere. Long-path and gray-line are not exotic — they're logged daily, and they'd be impossible on a static flat plane.

Sources: HF long-path & gray-line propagation [81]. Extends the skywave physics of Entry 46; tracks the terminator of Entry 44. → propagation data rows

• 1It isn’t only NASA — it’s everyone with a camera in space. Full-disk Earth imagers are run by the US (GOES), Japan (Himawari), Europe (Meteosat), Russia (Elektro-L), China (Fengyun) and India (INSAT), each returning a round disk roughly every 10 minutes. Russia and China have every geopolitical reason to expose an American hoax; instead their own satellites corroborate it.
• 2The whole sunlit disk, daily, from a million miles. NASA’s DSCOVR/EPIC, parked at the L1 point ~1.5 million km out, posts 12–22 public-domain images a day showing the entire sunlit face turning through a day — not one polished “hero” shot. Apollo’s 1972 Blue Marble, Galileo and the Lunar Reconnaissance Orbiter caught the full disk from yet other distances and angles.
• 3Fisheye is a red herring — and it cuts both ways. Lens distortion is identifiable: a fisheye bows straight lines, a rectilinear lens keeps them straight. The honest evidence isn’t a GoPro on a weather balloon (those wide lenses exaggerate the curve); it’s the convergent imagery above — plus the horizon dip you can measure yourself from a plane window (Entry 4), which needs no agency at all.

Sources: daily full-disk imaging, DSCOVR/EPIC [99]; the international geostationary fleet [100]. Pairs with the hardware of Entry 53 and the dip you can measure in Entry 4. → Curve data rows.

• GNSSYour location is satellite math. A phone fixes its position by timing signals from four or more GPS/GLONASS/Galileo/BeiDou satellites orbiting at ~20,200 km. The fix only works if those satellites are exactly where orbital mechanics puts them — and only if the receiver applies the relativistic clock correction (~38 µs/day, Entry 9). Disprove space and you disprove your maps app.
• DISHEvery satellite dish points at space. TV dishes aim at geostationary satellites 35,786 km above the equator. The fixed aim angle for your latitude/longitude is computed for an object parked in orbit; it works, repeatably, worldwide.
• ISSYou can see it yourself. The Space Station orbits at ~400 km, ~7.8 km/s, once every ~92 minutes. Apps predict its passes to the minute; it crosses the sky as a bright, fast point exactly on schedule — visible with your own eyes, no telescope.
• SENSORSThe other chips agree. The same phone holds a MEMS gyroscope (senses rotation — Entries 22, 18), a magnetometer (a compass reading Earth's dipole — Entry 27), an accelerometer, and often a barometer. The hardware used to reach TikTok is a small observatory confirming orbit, rotation, and a global magnetic field.

Sources: GNSS & satellite orbits [57] · GPS relativity [19] · phone sensors [22]. → Navigation data rows

• 1Naked-eye, on a published schedule. The ISS reaches magnitude −1 to −4 (occasionally −4.6), brighter than Venus and visible even from city centres. NASA’s “Spot the Station” lists every visible pass for your location — direction, time and maximum height — and passes always fall near dawn or dusk, when the station is sunlit against a dark sky. No flashing lights: that is how you know it is not a plane.
• 2Anyone can photograph it — independently of any agency. A phone on a tripod catches its streak; a 5-inch telescope resolves the modules and solar arrays. Astrophotographers plan ISS transits — the station’s silhouette crossing the Sun or Moon in under a second — from locations they choose, using public orbital data. These are private citizens, not NASA press releases.
• 3It behaves like an orbiting object on a globe. It circles every ~90 minutes at ~400 km and ~28,000 km/h, passing over about 90% of the world’s population, rising in the west, and vanishing the moment it enters Earth’s shadow. The station is a multinational operation (NASA, Roscosmos, ESA, JAXA, CSA) — the same “rivals would expose a hoax” logic as Entry 52.

Sources: NASA “Spot the Station” visibility & predictions [104]; amateur ISS transit photography [105]. Pairs with the satellite hardware of Entry 53 and the independent imagery of Entry 52. → Sky data rows.

• 1Two effects, opposite signs. Time dilation from the ~14,000 km/h orbital speed costs ~7 µs/day; the gravitational blueshift at ~20,200 km altitude adds ~45 µs/day. Net +38 µs/day — engineered into every satellite clock before launch.
• 2Failure is fast and large. Omit the correction and errors accumulate ~10 km per day; GPS would be useless for navigation within hours. The fix assumes orbits around a spherical, rotating Earth of a specific mass.
• 3Plus the Sagnac correction. Because Earth rotates, the timing must account for the receiver moving during signal transit (the Sagnac effect) — a direct consequence of a spinning globe, not a flat, still plane.

Sources: relativity in the Global Positioning System [89]. Sits with the satellites of Entry 53 and the laser-ranged LAGEOS of Entry 56. → Relativity data rows.

• 1A mirror you can shoot at. LAGEOS 1 (1976) and LAGEOS 2 (1992) are dense brass-and-aluminium balls ~60 cm across, covered in 426 corner-cube retroreflectors that send any incoming light straight back. Ground stations fire a laser, catch the returning photons, and the round-trip time gives the distance to a few millimetres. Anyone with the right gear can range them — the satellites are passive and carry no transmitter to fake.
• 2It reads the planet's shape and spin. The exact way these orbits wander reveals Earth's gravity field and its equatorial bulge (the oblate sphere of Entry 9), pins down the length of the day and polar motion, and tracks continents drifting centimetres a year. This is the backbone of satellite geodesy — the reference frame your phone's GPS is ultimately tied to (Entry 53).
• 3It even feels relativity. Einstein's general relativity predicts a spinning mass drags the local spacetime around with it. Laser ranging of LAGEOS 1 and 2 gave the first direct measurement of this frame-dragging from Earth's rotation — a tiny shift of the orbital planes of a few metres a year, matching the prediction. A flat, still plane has no spin to drag anything (Entry 20).

Sources: LAGEOS, satellite laser ranging & the frame-dragging measurement [67]. Confirms the oblate, rotating Earth of Entries 9 and 18; underpins the GPS of Entry 53. → geodesy data rows

• 1A real, distant, solid body. The round-trip light time runs 2.34–2.71 s as the Moon’s distance varies over its orbit (351,000–406,000 km); multiply by the speed of light and the lunar distance falls straight out. The echo comes from specific fixed corner-cube arrays — not from a glow.
• 2Anyone can check it. APOLLO uses a 3.5-m telescope and detects only a few returned photons per pulse (about 1 in 10¹⁷ survive the round trip); the Observatoire de la Côte d’Azur gets the same answer with different equipment. Independent replication is the mark of a real measurement, not a single-agency claim.
• 3It keeps doing science. Millimetre ranging has measured the 3.8 cm/yr recession, detected the Moon’s liquid core, mapped its libration (Entry 34), and tested the equivalence principle and the constancy of G — all of which require a genuine Earth–Moon two-body system.

Sources: lunar laser ranging & APOLLO [83]; lunar recession [84]. Kin to the laser-ranged LAGEOS of Entry 56 and the reflective regolith of Entry 32. → Moon data rows.

• 1Three dishes, 120° apart — because Earth turns. The DSN sits at Goldstone (California), Madrid (Spain) and Canberra (Australia), each roughly a third of the way around the planet. Before a distant spacecraft sets below the horizon at one site, the next rotates into view and takes over, giving unbroken contact. On a flat disc with everything visible at once, you would never need stations spread evenly around a sphere — the geometry is itself a rotating-globe experiment.
• 2The Voyagers, and a delay you can time. Launched in 1977, Voyager 1 is ~25 billion km away (~160 times the Earth–Sun distance) and crossed into interstellar space in 2012; its radio whisper, received on 70-metre DSN dishes, takes about 23 h 32 m one way and will hit a full light-day in late 2026. In 2023 a wrong command pointed Voyager 2's antenna away; Canberra sent an 18½-hour "interstellar shout," and 37 hours later (there and back) the craft answered. Those delays match the distance and the speed of light, every time.
• 3And rockets work better in vacuum. A rocket pushes off its own exhaust (Newton's third law — action and reaction), not off the air; with no atmosphere to fight, thrust is actually more efficient in space. The same network flies the Parker Solar Probe to the Sun (Entry 45) and tracked every crewed mission. The hardware, the light-delays and the geometry all hang together — and all require real distance through real space.

Sources: NASA Deep Space Network [68] · Voyager mission distance & light-time [69] · Parker Solar Probe [70]. See also satellites (Entry 53), radio propagation (Entry 46) and timing the Sun (Entry 45). → deep-space data rows

• 1One specific altitude does it. At 35,786 km above the equator an orbit takes exactly one sidereal day (23 h 56 m), matching Earth's spin, so the satellite stays fixed in the sky. Arthur C. Clarke described it in 1945; today hundreds of satellites occupy that ring. No other altitude gives a motionless bird — the number falls straight out of orbital mechanics on a spinning globe.
• 2Every dish agrees, and points at the sky. Installers aim dishes using latitude/longitude look-angle tables; the farther north you are, the lower toward the southern horizon the dish tilts (northern hemisphere). All those lines of sight, from every continent, intersect the same equatorial arc tens of thousands of km up — not at any ground tower. A flat Earth has no geometry that makes those angles consistent.
• 3Ground transmitters can't fake it. A dish has a narrow beam aimed at empty sky above the equator; block its view of that arc and it fails, while terrestrial signals from the "wrong" direction are rejected. The latency through a geostationary hop (~0.25 s round trip) also matches a ~72,000 km up-and-down path. It's the same satellite reality as GPS (Entry 53), tracked by networks like the DSN (Entry 58).

Sources: geostationary orbit & the Clarke belt (35,786 km) [82]. Same satellite reality as Entry 53; tracked via networks like Entry 58. → satellite data rows

• 1What the treaty actually says. Signed in 1959, in force 1961, now some 56–58 nations (29 with decision-making "Consultative" status), covering everything south of 60°S. It demilitarises the continent, bans weapons and nuclear tests, freezes territorial claims, and — crucially — guarantees freedom of scientific investigation and the free exchange of results. It opens the continent to researchers; it forbids armies, not visitors.
• 2People reach it, cross it, and circle it. The South Pole was reached in 1911 (Amundsen) and has held a permanently staffed station (Amundsen–Scott) since 1956, resupplied by air. The continent was crossed surface-to-surface in 1957–58 (Fuchs & Hillary) and many times since, is routinely circumnavigated by ship, and is overflown (Entry 24). Its coastline runs ~18,000 km and its area is ~14 million km² — the geometry of a continent, not the tens-of-thousands-of-km rim a disc-edge wall would need.
• 3The "ice wall" is a coast. Antarctica's floating ice shelves end in cliffs — the Ross Ice Shelf's "Great Ice Barrier" stands ~15–50 m above the sea. You can land, climb onto the shelf, continue up onto the ~2 km-thick ice sheet, cross the pole, and descend to the ocean on the opposite side. A wall that encloses everything cannot be walked over and past.
• 4What is truly unexplored — and why it isn't "lands beyond." Real frontiers remain: only ~27% of the deep seafloor is mapped to modern resolution (2025), and the bedrock under Antarctica's ice hides buried mountain ranges (the Alps-sized Gamburtsevs) and sealed lakes (Lake Vostok, under ~4 km of ice). But these are under water and under ice, not past an edge. Satellite gravimetry and geodesy fix the elevation and position of the entire surface; the area budget closes with no room for an undiscovered continent.

Sources: Antarctic Treaty [59] · Antarctic geography & exploration [60] · seafloor & subglacial mapping [61]. See also the satellites overhead (Entry 53). → Antarctica data rows

• 1Perspective, not divergence. The Sun is ~150 million km away, so the rays reaching Earth are parallel to a tiny fraction of a degree. Cloud-gap shadows make them visible; perspective makes parallel lines appear to radiate from the Sun’s direction.
• 2The reconvergence test. Follow the rays past the zenith and they appear to meet again at the antisolar point — directly opposite the Sun. Parallel lines have two vanishing points; a small local source has none. Photographs and views from the ISS confirm the rays are uniform and parallel.
• 3It backfires. Offered as evidence of a near Sun, crepuscular rays — once anticrepuscular rays are accounted for — only make sense for a Sun effectively at infinity. The exhibit refutes the very claim it is meant to support.

Sources: crepuscular & anticrepuscular rays [86]. Complements the Sun’s constant angular size (Entry 40) and its measured distance (Entry 45). → Sun data rows.

• 1Magnification is not elevation. A 2-metre-high viewpoint puts the sea horizon about 5 km away (distance in km ≈ 3.57 × √height in metres). Past that, a ship’s hull is blocked by the curve. Zooming enlarges the visible superstructure but cannot lift your eye-line over the water; the hull stays cut off at the same waterline, just larger.
• 2“Coming back” is refraction — which needs a curve. On days with a strong temperature inversion (warm air over cold water) light bends downward and follows the surface, lifting hidden parts into view. That bending is real and measurable — and it is only necessary because the surface curves away. A flat sea would need no such rescue.
• 3The honest version of the test. Note the ship’s height, your eye height and the distance; predict how much hull should be hidden (use the curvature calculator in Entry 3), then look. Under steady air it matches the globe to within the refraction margin. The “gotcha” becomes a measurement.

Sources: terrestrial refraction & looming [90]. Built on the horizon geometry of Entry 1 and the photography test of Entry 3. → Curvature data rows.

• 1Tops visible, bases hidden — that is the curve. Chicago is about 85 km (53 miles) across the lake. A 6-foot observer sees ~3 miles to the horizon; even from a 250-foot dune, ~20 miles. The Willis Tower (442 m / 1,450 ft) is tall enough for its top to clear the curve from ~65 miles, but its base is geometrically blocked. Photos show towers rising out of the water with their lower floors missing — the signature of a sphere.
• 2The full skyline is a mirage — and mirages need a curve. When warm air sits over the cold lake, light ducts downward and lifts hidden buildings (sometimes flipping them) into view; meteorologists document this routinely. A mirage is real refraction, not an illusion — and there is nothing to bend “back into view” unless the surface curves away in the first place.
• 3Every famous “impossible” sighting resolves this way. Put observer height and target height into the horizon formula (Entry 1), allow for standard refraction, and the distant lighthouses, mountains and skylines fit the globe. The exhibit meant to disprove curvature ends up measuring it.

Sources: long-distance observation, refraction & looming [90]. Uses the horizon geometry of Entry 1 and the curvature calculator in Entry 3. → Curvature data rows.

• 1Lake Pontchartrain’s tower line is a ready-made curvature experiment. A power transmission line crosses the lake in a straight run of ~16 miles, its pylons identical in height and evenly spaced (~287 m apart). Shot end-on with a telephoto (as the photographer “Soundly” did from 2017), the bases of the farther towers sit progressively lower and vanish behind the water — and the line’s vanishing point lands above the horizon, not on it. Flat perspective predicts the opposite: equal towers shrinking toward a point on the horizon, all bases on one line.
• 2The salt flats are “level,” and level means curved. Bonneville’s salt is laid down by evaporating water, so its surface is an equipotential — the same curved “level” as the sea. The National Geodetic Survey finds only ~7.9 inches of height variation across the whole flat, yet over a 10-mile span a sphere’s geometric drop is ~67 feet. You don’t see a 67-foot cliff because you are standing on the curve; the far end simply falls below your local horizon — and from the hills above Wendover, a long lens shows Interstate 80 bending over the rim.
• 3“It looks flat” is about scale, not shape. From a low vantage the curve per mile is gentle and the eye has nothing to measure against, so a salt pan or a calm lake reads as flat — exactly as Entry 4 (the horizon dip) and Entry 6 (water’s level) explain. Raise the viewpoint, stretch the sightline, or add a zoom lens, and the same “flat” places reveal the bend.

Sources: Lake Pontchartrain power-line curvature [101]; Bonneville Salt Flats levelness vs. curvature [102]. Kin to the Bedford Level (Entry 2) and the ship that drops hull-first (Entry 62). → Curve data rows.

• 1Half lit, always. A distant Sun lights one hemisphere of a sphere at a time, so roughly 50% of Earth is in day and 50% in night at every moment, split by a crisp terminator. A nearby lamp over a disc would cast a bright circle that fades outward — never a clean half-and-half with a sharp edge.
• 2Antipodes run ~12 hours apart. Earth turns 360° in 24 hours — 15° per hour — so each ~15° of longitude is about an hour later. Opposite sides of the globe sit ~12 hours apart: your midday is your antipode’s midnight, consistently, all year. On a flat disc the far rim would never be reliably opposite in time.
• 3You can verify it from your couch. Video two people on opposite sides of the world and note their local times and whether it is light. The pattern only closes on a rotating sphere — it is the day/night model of Entry 44 sitting in your call log.

Sources: time zones & Earth’s rotation [91]. The mechanism is the terminator model of Entry 44; the signal timing echoes the latency of Entry 49. → Sun data rows.

• 1Both poles, opposite states, same day. Above the Arctic Circle (66.56° N) the Sun stays up 24 hours at the June solstice; at that exact moment, below the Antarctic Circle (66.56° S) it never rises. In December it reverses. Two opposite, simultaneous behaviours rule out a single lamp over a single disc.
• 2The Circle latitude is the tilt, exactly. 66.56° = 90° − 23.44°, Earth’s axial tilt. The midnight Sun begins precisely where the geometry of a tilted, spinning sphere says it must — a number with no meaning on a flat plane.
• 3The Sun circles without setting — as seen from a tilted pole. Near a pole in summer the Sun does loop around the horizon all day, but at a steady height, because you are standing near the top of a tilted ball facing the distant Sun. It is the tilt and the sphere together (see the analemma, Entry 39, and the terminator, Entry 44).

Sources: midnight Sun & polar night [92]. Follows from the axial tilt behind the analemma (Entry 39) and the day/night terminator (Entry 44). → Sun data rows.

No lab required. Each of these uses household items or a phone, and each is hard to reconcile with a flat, non-rotating Earth.

• 1Watch a ship leave. With binoculars from a shoreline, a departing ship loses its hull before its mast — the bottom hides behind the bulge first. (The hidden-base effect, Entry 3.)
• 2Polaris = your latitude. Measure Polaris's angle above the northern horizon (protractor + plumb line, or ~10° per fist at arm's length). It equals your latitude — and changes as you travel north or south, which a flat dome can't do. South of the equator Polaris is gone entirely.
• 3Two sunsets. On a flat-horizon beach, watch the Sun fully set while lying down, then stand up quickly — you'll see it set a second time. Gaining a few metres pushes your horizon back; only a curve does that. Time the gap and you can even estimate Earth's size.
• 4DIY Eratosthenes. Two people a few hundred km apart (north-south) measure a vertical stick's shadow at the same moment (coordinate by phone). The shadow angles differ; the difference relative to the distance gives Earth's circumference. A flat Earth under a distant Sun would give identical shadows.
• 5Call across the world. Phone someone a third of the way around the globe: midday for you, night for them. One flat disc under one Sun cannot be day and night simultaneously at the spacings observed.
• 6Watch a lunar eclipse. Earth's shadow on the Moon is always a circle's edge — every eclipse, every orientation. Only a sphere casts a round shadow from every angle.
• 7Look south. From the Northern Hemisphere you never see σ Octantis or the deep southern sky; travel south and new constellations rise while northern ones set — two distinct celestial hemispheres.
• 8A pendulum or a phone gyro. A long pendulum's swing plane slowly turns (Foucault, Entry 18); even a phone's MEMS gyroscope, logged carefully and averaged, can reveal the ~15°/hr rotation signature at its noise floor.

Methods draw on Entries 1, 3, 18 & 22 and their sources. → Curvature & pole-star data rows