# Where the numbers come from Everything EarthLens draws comes out of `modelEarth()` in `index.html`: a zero-dimensional energy-balance climate model with ice-albedo and water-vapour feedback, sitting on top of keyframe curves for CO₂, O₂, methane, vegetation, biosphere, continental growth and supercontinent assembly. This file says what each piece rests on, and — the part that matters more — where the model is knowingly wrong. **Nothing here is invented history. Where a number is a guess, it says so.** Citations are to the standard papers for the shape of each curve. The keyframe values are my own hand-fitting to those shapes, not transcriptions from any one table, and the deep-time ones carry error bars far wider than the tidy numbers in the app suggest. --- ## The Sun `L(t) = 1 / (1 + 0.4 (1 − t/t_now))`, with the Sun's present age 4570 Myr. - Gough, D.O. (1981), *Solar interior structure and luminosity variations*, Solar Physics 74, 21. This is the standard main-sequence approximation and is used essentially unchanged. - It gives 0.72 at Earth's formation — the faint young Sun — and 1.10 at +1 Gyr. **Where it cheats.** The Sun's first ~50 Myr were pre-main-sequence and this formula does not describe them, so solar age is clamped to 100 Myr. It does not matter: at that point the surface temperature is set by accretional heat, not sunlight. ## Climate Surface temperature is solved by relaxation: T = T_e + G(CO₂) + G(H₂O) + G(CH₄) + accretional heat T_e = (S₀ L (1−A) / 4σ)^¼, S₀ = 1361 W/m² Calibrated so the present day returns exactly: albedo 0.300, T_e 254.6 K, greenhouse 33.4 K split 7 K to CO₂ and 26.4 K to water vapour, T = 288 K. - Present albedo 0.29–0.30: Stephens et al. (2015), *The albedo of Earth*, Rev. Geophys. 53, 141; Trenberth, Fasullo & Kiehl (2009), *Earth's global energy budget*, BAMS 90, 311. - The ~1/4 : ~2/3 split of the greenhouse between CO₂ and water vapour: Schmidt et al. (2010), *Attribution of the present-day total greenhouse effect*, JGR 115, D20106. - CO₂ forcing logarithmic in concentration: Myhre et al. (1998), GRL 25, 2715. - Equilibrium sensitivity: IPCC AR6 WG1 Ch.7 gives 2.5–4 K per doubling, best estimate 3 K. This model lands near **4 K**, the upper half of that range — a deliberate choice, because a less sensitive model will not reproduce deep-time swings without implausible CO₂. **Where it cheats.** Above about 1% CO₂ the plain logarithm badly underestimates forcing, because a thick CO₂ atmosphere pressure-broadens its own absorption bands. A super-logarithmic term is bolted on above 10,000 ppm (see Byrne & Goldblatt (2014), *Radiative forcings for 28 potential Archean greenhouse gases*, Climate of the Past 10, 1779, for why the correction is needed). Its exponent is fitted by eye, not derived. ### Ice and the snowball bifurcation Ice cover against temperature is piecewise: gentle either side of the present day (10% at 288 K), then steep below 272 K, where it runs to a full snowball by 263 K. That kink is not decoration. It is the ice-line instability — once the ice edge reaches roughly 30° latitude the albedo feedback outruns the restoring forces and the planet freezes solid. Keeping the feedback gain near 0.65 in the gentle region and far above 1 in the steep one is what lets the model give a 4.5 K glacial-interglacial swing *and* a Snowball Earth from the same equations. - Budyko (1969), Tellus 21, 611; Sellers (1969), J. Appl. Met. 8, 392 — the original energy-balance models and the ice-line instability. - Hoffman & Schrag (2002), *The snowball Earth hypothesis*, Terra Nova 14, 129. - Hoffman et al. (2017), *Snowball Earth climate dynamics and Cryogenian geology–geobiology*, Science Advances 3, e1600983. **Where it cheats.** The model has no memory, so it has no hysteresis: it always relaxes from 288 K and therefore finds the warm branch whenever one exists. A real snowball, once entered, persists until CO₂ builds to percent levels. Here that persistence is written into the CO₂ keyframes instead — which is why CO₂ stays low through the glaciations and spikes at the deglaciations. ## Carbon dioxide Keyframed in ppm, interpolated geometrically. - **Phanerozoic** (541 Ma → now): Berner (2006), *GEOCARBSULF*, Geochim. Cosmochim. Acta 70, 5653; Royer et al. (2004), *CO₂ as a primary driver of Phanerozoic climate*, GSA Today 14, 4; Foster, Royer & Lunt (2017), *Future climate forcing potentially without precedent in the last 420 million years*, Nature Communications 8, 14845. The Carboniferous minimum (~400 ppm, the coal forests drawing carbon down), the end-Permian spike (~2500 ppm, Siberian Traps), the Cretaceous hothouse (~1600 ppm) and the Cenozoic decline all come from these. - **Pleistocene**: 190 ppm at the Last Glacial Maximum, ~280 ppm pre-industrial, 425 ppm today — EPICA Dome C ice core, Lüthi et al. (2008), Nature 453, 379; NOAA GML for the present value. - **Proterozoic and Archean**: order-of-magnitude only. Kasting (1993), *Earth's early atmosphere*, Science 259, 920; Sheldon (2006) and Driese et al. (2011) on paleosol constraints. **Where it cheats, and it matters.** - The Cryogenian onsets are set to ~120–150 ppm. That is lower than most reconstructions. Low CO₂ is the standard trigger (basalt weathering after the Franklin large igneous province), but the absolute value at the onset is very poorly constrained, and these keyframes are **tuned to make the model cross its own snowball threshold**, not read off a proxy. - The Hirnantian (end-Ordovician) glaciation comes out as only a mild cooling. The real one was driven largely by having a continent sitting on the south pole, and this model has no continental-position effect at all. It is honestly weak here. - Archean values are chosen jointly with the methane term below, because the two trade off against each other and neither is independently well constrained. ## Methane and the two oxygenations An extra greenhouse term, in kelvin, standing for methane and the other reduced gases that an anoxic atmosphere sustains: ~10 K through the Archean, collapsing at the Great Oxidation Event, recovering to ~4.5 K through the low-oxygen Proterozoic, declining again with Neoproterozoic oxygenation. - Pavlov et al. (2000), *Greenhouse warming by CH₄ in the atmosphere of early Earth*, JGR 105, 11981 — the Archean methane greenhouse as the answer to the faint young Sun. - Catling, Zahnle & McKay (2001), Science 293, 839; Haqq-Misra et al. (2008), Astrobiology 8, 1127. - Bekker et al. (2004), *Dating the rise of atmospheric oxygen*, Nature 427, 117 — the GOE at ~2.45–2.32 Ga. - Kopp et al. (2005), PNAS 102, 11131 — oxygenation destroying the methane greenhouse and causing the Huronian glaciations. This is why, in the app, moving the time dial from 2445 to 2395 Ma drops the planet from +9 °C to −54 °C. That causal chain is real, and it is the thing I most wanted the app to get right. **Where it cheats.** The methane curve is expressed directly in kelvin of warming rather than in mixing ratio, so it bypasses the radiative transfer entirely. The magnitudes are in the range those papers give; the shape between the two cliffs is my interpolation. ## Oxygen Keyframed in percent by volume: a trace before 2.45 Ga, a step at the GOE, a long low Proterozoic plateau, the Neoproterozoic rise, ~32% at the Carboniferous peak, the end-Permian crash, 20.95% today. - Lyons, Reinhard & Planavsky (2014), *The rise of oxygen in Earth's early ocean and atmosphere*, Nature 506, 307. - Holland (2006), *The oxygenation of the atmosphere and oceans*, Phil. Trans. R. Soc. B 361, 903. - Berner (2006), GEOCARBSULF, for the Phanerozoic curve and the Carboniferous maximum. **Where it cheats.** Future oxygen is simply tied to vegetation cover — if the land biosphere dies, O₂ is drawn down to zero. The real drawdown by oxidation of crust and organics takes a few million years and has its own chemistry. This is a one-line proxy. ## Water, and losing it Oceans ramp in between 4.54 and 4.26 Ga; thereafter constant until the deep future. - Wilde et al. (2001), *Evidence from detrital zircons for the existence of continental crust and oceans on the Earth 4.4 Gyr ago*, Nature 409, 175; Valley et al. (2002), Geology 30, 351. - Zahnle et al. (2007), *Emergence of a habitable planet*, Space Science Reviews 129, 35. Future loss is a ramp between +200 and +900 Myr, scaled by how far the effective solar luminosity has risen past 1.0 — so dimming the Sun with the solar dial prevents it, which is the physically coherent behaviour and is what makes the flourishing ending reachable. - Kasting, Whitmire & Reynolds (1993), *Habitable zones around main sequence stars*, Icarus 101, 108. - Leconte et al. (2013), *Increased insolation threshold for runaway greenhouse processes on Earth-like planets*, Nature 504, 268 — the moist greenhouse threshold near 1.1 S₀. - Wolf & Toon (2015), *The evolution of habitable climates under the brightening Sun*, JGR Atmospheres 120, 5775. **Where it cheats.** Real water loss is driven by how hot the planet actually is; here it is driven by time and luminosity, which is a proxy for the same thing but cannot feed back. Estimates of when Earth loses its ocean range from about +1 Gyr to +2 Gyr, and this model is at the pessimistic end of that. The water-vapour greenhouse is keyed to **ocean coverage**, not total volume, because the air stays near saturation while any sea remains. Keying it to volume — the first thing I tried — let a half-drained planet freeze itself, which is backwards. ## The end of plants Vegetation is suppressed below ~150–200 ppm CO₂. - Caldeira & Kasting (1992), *The life span of the biosphere revisited*, Nature 360, 721 — the original calculation that CO₂ starvation, not heat, ends the land biosphere first. - Lovelock & Whitfield (1982), *Life span of the biosphere*, Nature 296, 561. - Franck et al. (2000), *Reduction of biosphere life span as a consequence of geodynamics*, Tellus B 52, 94. - O'Malley-James et al. (2013), *Swansong biospheres*, International Journal of Astrobiology 12, 99. Vegetation also falls when the continents gather. Rain comes off the sea, so the middle of a large landmass does not get any, and when a supercontinent exists a great deal of land is a long way from a coast. The model takes 22% off global vegetation at full assembly, tapering to nothing when the continents are dispersed, which is why Pangaea here is a desert with a green fringe. - Parrish (1993), *Climate of the supercontinent Pangea*, Journal of Geology 101, 215. - Tabor & Poulsen (2008), Palaeogeography Palaeoclimatology Palaeoecology 268, 293 — the aridity of the Pangaean interior in the Permian and Triassic. **Where it cheats.** C4 plants can fix carbon down to ~10 ppm and would outlast C3 plants by hundreds of millions of years. The model has one plant type with one threshold, so it kills the biosphere earlier than a careful treatment would. The 22% is a round number chosen to match the qualitative picture, not derived; it also docks the Carboniferous, whose coal forests were real but sat in a coastal equatorial belt the model cannot distinguish from the interior. ## Life and the extinctions A biosphere index, 0–1, keyframed with the Cambrian explosion and the five big crashes: end-Ordovician (445 Ma), Late Devonian (372 Ma), end-Permian (252 Ma), end-Triassic (201 Ma), end-Cretaceous (66 Ma). - Raup & Sepkoski (1982), *Mass extinctions in the marine fossil record*, Science 215, 1501. - Sepkoski (2002) compendium; Alroy et al. (2008), Science 321, 97 — sampling-corrected diversity. - Erwin (2006), *Extinction: How Life on Earth Nearly Ended 250 Million Years Ago*. **Where it cheats.** It is one number standing for all of biodiversity, productivity and reach. It has no units and should be read as a mood, not a measurement. ## Continents Ten cratonic blocks, each interpolating between a dispersed position (roughly where that craton is today) and a gathered position relative to a drifting assembly centre, under an assembly parameter keyframed to the supercontinent cycle: Vaalbara ~3.1 Ga, Kenorland ~2.7 Ga, Columbia/Nuna ~1.6 Ga, Rodinia ~1.1–0.75 Ga, Pannotia ~0.6 Ga, Pangaea ~0.3–0.2 Ga, dispersed today, the next one at ~+250 Myr. Continental area grows from nothing to today's over the first two billion years. - Nance, Murphy & Santosh (2014), *The supercontinent cycle: a retrospective essay*, Gondwana Research 25, 4. - Merdith et al. (2021), *Extending full-plate tectonic models into deep time*, Earth-Science Reviews 214, 103477 — the 1 Ga-to-present reconstruction this is a caricature of. - Hawkesworth et al. (2010) and Cawood et al. (2013) on continental crustal growth. - Scotese (2003) *Pangaea Ultima*; Davies, Green & Duarte (2018), Global and Planetary Change 166, 184 — the candidate future supercontinents. Farnsworth et al. (2024), *Climate extremes likely to drive land mammal extinction during next supercontinent assembly*, Nature Geoscience 17, 724. ### Where the continents actually were From 0 to 1000 Ma the continents are not modelled at all. They are **Merdith et al. (2021)**, a published full-plate reconstruction, fetched once from the **GPlates Web Service** (`gws.gplates.org`) and rasterised here into 27 equirectangular masks. CC-BY; the credit is in the footer of the page. - Merdith, A. S., et al. (2021), *Extending full-plate tectonic models into deep time: Linking the Neoproterozoic and the Phanerozoic*, Earth-Science Reviews 214, 103477. - Müller, R. D., et al. (2018), *GPlates: Building a virtual Earth through deep time*, Geochemistry Geophysics Geosystems 19, 2243. The slices are stored as bilevel masks — the whole billion years is 36 KB — and the browser rebuilds a signed distance field from each at load. **The app interpolates the distance fields, not the masks.** That distinction is the difference between a coastline that glides between two ages and one continent dissolving into another. The supercontinent cycle is nowhere in the code. It falls out of the data: measuring how much of the land sits in its largest connected mass gives 97–98% at 135–160 Ma (Pangaea), 21% at 620 Ma (maximum dispersal after Rodinia breaks up), and 71% again at 1000 Ma (Rodinia). **Where it cheats, and what the data cannot say.** - These polygons trace **continental margins, not shorelines**, so at face value they put 35.2% of the present-day surface above water against Earth's 29.2%. The contour is stepped 1.2 texels (about 190 km) inland to fix that, and the same offset is applied at every age. - **Coverage thins going back.** Land area in the reconstruction falls from 35% today to 20% at 1000 Ma. Some of that is real — continents did grow — but some of it is simply that less is known about early Neoproterozoic coastlines. The app cannot separate the two, and treats the reconstruction's land as the land. - The rasters are 256×128, so a slice resolves about 150 km. Coastline detail finer than that is the model's noise, not the reconstruction's. - Slices are 10–80 Myr apart. Between them continents move in a straight interpolation, which is not how plates move. - Lowering the land area at 635 Ma took the **Marinoan** glaciation below the model's snowball threshold, so that keyframe's CO₂ was lowered until it tripped again — the same tuning already declared for the Sturtian, and the same caveat applies: it is tuned to produce the glaciation the record requires, not measured. - Nothing reconstructs the **future**, either. Past the present day the model's own blocks take over, fading in over the first 60 Myr. Before 1000 Ma, and in the future, the continents are the model's own ten cratonic blocks: The sphere is bent by a low-frequency displacement before distances to the blocks are measured. A block is a cone and a cone's contours are circles, so without that every continent comes out an oval however much the coastline is crenellated; the bend stretches the same cones into irregular masses with bays, capes and narrow necks. It is fixed in space while the blocks drift through it, which means a continent slowly changes shape as it travels — not the reason real ones do, but not a wrong thing to show either. **Where it cheats, loudly.** These are blobs, not continents. The assembly parameter is a single scalar when real supercontinents assemble piecewise over hundreds of millions of years; blocks drift through each other; the coastlines are noise. The picture is for showing *that* the continents move and gather, never *where* anything was. ## What the picture adds, and what it does not Terrain relief and coastline detail are fractal noise. They carry no information: there is no claim that a mountain range was ever there. What the model does set is how much of each thing there is, and the picture is tied to the readouts rather than drawn beside them: - Ice in the drawing is placed at the latitude that makes its area equal the ice fraction the readout reports, so the two cannot disagree. Land ice is pushed a little further equatorward than sea ice, which is how it sits on the real planet, and the two still average to the model's number. Snow on high ground is extra, and is cosmetic. - Biome colour — rainforest, temperate, boreal, tundra, steppe, desert — is chosen from the model's own zonal temperature profile with a lapse rate applied for altitude, and the green/bare split follows the model's vegetation fraction, graded across a few hundred kilometres rather than drawn as a line. - Where the green sits is set by distance from the sea as well as by latitude: coasts are wet, interiors are dry, written as a redistribution so the green area still roughly matches the fraction the model reports. The same effect is in the model itself, so the readout agrees. - Mountains are built into belts a little way in from the margins rather than scattered over interiors, because the Andes, the Rockies and the Himalaya are belts and what lies behind them is plain. Which margin gets one is noise; that any does is not. - The sea's own colour answers to the biosphere. Plankton follows the shelves, the upwellings and the cold water and leaves the mid-ocean gyres alone, which is why the open Pacific is the bluest water on the planet; the greening scales with the model's biosphere index, so an Archean ocean or one the dials have sterilised is pure blue, with no green at the margins at all. - **Clouds are not drawn** (removed 25 October 2026, at Maria's call). They were modelled in some detail — banded on the real circulation, cellular where convection drives them, wound into commas by depressions turning the right way for their hemisphere — and the whole layer came out, because the subject of the picture is the land and the weather was covering it. They remain in the physics: the `CLOUD` term supplies most of the planet's albedo in the energy balance, and deleting that would warm the model by about thirty kelvin. So the Earth here is cloudless in the way a map is cloudless, not in the way a planet without an atmosphere is. - Islands are in the picture only. Arcs and archipelagos trail the margins because continents without them read as closed ovals, but they are about one part in a hundred of Earth's land — below what the model's block geometry resolves — so the land and ocean percentages in the readouts ignore them. The same goes for the inland lakes, which sit in basins away from the coast where there is rain to fill them: at this distance rivers are mostly too narrow to see, but the large lakes are not, and land with no water on it anywhere looks wrong. - Silt is drawn where rivers would reach the sea, brown against the blue in the shallows. The Amazon, the Ganges and the Yellow River are all visible from orbit by their plumes. - The colour of the limb glow follows atmospheric composition: blue while there is water and normal air, orange once CO₂ is percent-level, grey when the air is thin. - Scattered skylight is added on top of the surface rather than blended over it, and it takes contrast away as well as adding blue — the deeper the air on the line of sight, the more of what reaches the eye was scattered on the way rather than reflected off the ground. Land seen through an atmosphere is muted, and leaving that out is what keeps a rendered planet looking painted. - The sun's glint on water is spread wide rather than drawn as a point. A mirror would give a point; wave slopes scatter the reflection over ten or fifteen degrees, which is why the glint in a photograph of the whole disc is a soft silvery patch (Cox & Munk (1954), *Measurement of the roughness of the sea surface from photographs of the sun's glitter*, JOSA 44, 838). ### The palette, measured rather than guessed The surface colours are not invented. They were sampled from NASA's Blue Marble composite (`land_shallow_topo_2048.jpg`, Reto Stöckli / NASA Earth Observatory, public domain), taking the median of a small patch at each site. The image was used once, as a reference, and discarded; it is not shipped, and the app still downloads nothing. | site | measured | |---|---| | open ocean (Pacific, Atlantic, Indian — all identical) | `#0b0a32` | | shallow carbonate shelf (Bahamas) | `#0f717b` | | rainforest (Amazon / Congo) | `#21380b` / `#2b3e0a` | | temperate forest (eastern USA / Europe) | `#2e470c` / `#3a4e15` | | boreal forest (Siberia / Canada) | `#29400b` / `#2a3b0f` | | tundra (northern Siberia) | `#545029` | | savannah (Sahel) / grassland (Great Plains) | `#c0ad69` / `#6a6c3a` | | desert (Sahara / Gobi / western Australia) | `#fdf3ac` / `#cdab85` / `#845e41` | | bare rock (Tibet / altiplano) | `#b39464` / `#a98662` | | ice (Greenland / East Antarctic plateau) | `#ffffff` / `#bdb5a9` | Three things came out of that, and all three had been wrong: 1. **Open ocean is almost black.** `#0b0a32` is a dark indigo in which green is *lower* than red. The sea looks blue in a photograph because of the atmosphere above it, not because the water is that colour. Blue Marble is a surface composite with the air taken out, which is why it is the right thing to calibrate an unlit surface map against — and it is why the renderer now adds scattered skylight as light on top of the surface rather than as a wash blended over it. 2. **Vegetation is dark and olive**, with blue around 10 out of 255. The greens here had been roughly twice as bright and four times as blue. 3. **Closed canopy is the darkest thing on the land surface**, so density now darkens vegetation where it used to lighten it. **Where it cheats.** The reference is one composite of one month, so the northern forests are in leaf and there is little sea ice in it. Nothing was measured for the sea floor, for cloud, or for anything that is not present-day Earth; those remain judgement. ### The present day is a photograph Within about a million years of now, and only while the dials are near their defaults, the land and sea are not drawn from the model at all: they are NASA's Blue Marble (public domain), the same 1024×512 equirectangular texture the sibling Hello World AI puts on its three.js globe. It is embedded in the page rather than linked, because a page opened from the filesystem cannot read pixels back out of a separate image file — the canvas is tainted and `getImageData` throws. It is used only for what it actually records, and three things are deliberately kept from it: - **Its ice.** The texture's polar caps are crude static white bands reaching to about 60°. The model's ice is used instead, so the carbon dioxide dial still moves the ice line at the present day, which is the first thing anyone tries. - **Its relief.** The model's slope map belongs to the model's continents, so it is faded out as the photograph takes over; otherwise mountains would be shaded where the photograph shows sea. - **Everything else.** Lighting, glint, the air and all the readouts stay the model's. How much it is believed falls off with time (gone by 12 Ma) and with how far the dials have moved the planet from today — measured on temperature, vegetation and water, so a warmed or greened or dried Earth is drawn by the model even at age zero. The caption says which you are looking at. The app icons are the same photograph: `ops/make_icons.py` pulls that data URI back out of `index.html`, projects it orthographically and lights it. So the icon on a phone's home screen is NASA's Earth, not an impression of one, and it is public domain for the same reason the texture is. **Where it cheats.** The hand-over is a cross-fade between two unrelated geographies, so between about 2 and 10 Ma the real continents dissolve into the model's. Real continents barely move in that time; nothing about the dissolve is a claim about drift. ### The sea floor The bathymetry has the right *shape* for an ocean basin — a narrow shelf of varying width, a continental slope, an abyssal plain around 5–6 km down, and a mid-ocean ridge system rising a couple of kilometres back out of it well away from any continent — but none of it is a real sea floor. It is noise shaped by distance from the nearest continent. What is physical is the water above it. Each colour is attenuated over the path length through the water, with scale depths of roughly 42 m for red, 115 m for green and 240 m for blue, which is the usual ordering for clear ocean and the reason a sandy shelf reads turquoise. Below a few hundred metres the floor contributes nothing and the colour is the water's own, which is then shaded by the sea-floor relief — so the picture carries the shape of the basin the way a bathymetric composite does, rather than the way a photograph would. - Jerlov (1976), *Marine Optics*, for the ordering and rough magnitude of the attenuation. - Depths after the standard hypsometric curve: shelf ~130 m, abyssal plain ~4–6 km, ridge crests ~2.5 km. **Where it cheats.** The vegetation pattern that decides which parts of the land are green is a fixed function of latitude and longitude, so the deserts sit in the same places in every era. Relief shading uses a fixed exaggeration, and the mountains and ridges are far higher, relative to the planet, than Earth's are. There are no trenches, no rivers and no lakes. ## Sea level Computed from four terms, referenced to the present so that today reads zero: ice volume (capped — in a snowball most of the water is still in the basin as sea ice, not piled on land), ocean volume, thermal expansion, and ridge volume from how dispersed the continents are. - Haq & Schutter (2008), *A chronology of Paleozoic sea-level changes*, Science 322, 64. - Miller et al. (2005), *The Phanerozoic record of global sea-level change*, Science 310, 1293 — including the Cretaceous highstand. - Clark et al. (2009), Science 325, 710 — LGM lowstand near −125 m. The model gives about −175 m, which is too deep. Shoreline position on the globe is not taken from this figure. It is solved separately: the height field is sampled on an equal-area grid, sorted once, and the shoreline is placed at whatever height makes the ocean cover the fraction that the available water can cover. Draining water only bares land where there is continental crust to bare — which is why a half-filled Hadean ocean is still a global one. ## Zonal temperature `T(lat) = T_mean + 15 − 45 sin²(lat)`, giving about 303 K at the equator and 258 K at the poles for a 288 K mean, and integrating back to the mean exactly. - Standard textbook profile; e.g. Hartmann, *Global Physical Climatology*, 2nd ed. (2016), Ch. 2. This is what places the ice edge in the drawing, and it is tied to the same ice fraction the readout shows, so picture and number cannot disagree. **Where it cheats.** One fixed profile for 4.5 billion years. The real equator-to-pole gradient is much flatter in hothouses and steeper in icehouses. --- ## The short list of things I would not defend 1. The continents are a caricature. Read nothing geographic off them. 2. Cryogenian CO₂ is tuned to trip the model's snowball, not measured. 3. The Hirnantian glaciation barely registers, because there is no continental-position effect. 4. Climate sensitivity is ~4 K per doubling — the high side of the accepted range. 5. The Hadean and Archean come out warm (22–30 °C). Defensible, but contested. 6. Future ocean loss is pinned to time and luminosity rather than to temperature, and is at the pessimistic end of published estimates. 7. One plant type, one CO₂ threshold, so the biosphere dies earlier than C4 plants would allow. 8. The LGM lowstand is about 50 m too deep. 9. There is no hysteresis anywhere, so no state depends on how the planet got there. 10. Mountains, coastlines and the sea floor in the picture are noise. Only their quantities mean anything.