Planet Generation — Astral Polity

Each of these is in the solver, with the citation in the source. Where a coefficient was chosen, the published range it was chosen from is recorded beside it.

Symbol	Quantity	Dimension	Units
z	ground-surface elevation	L	m
t	time	T	yr
U	rock uplift rate, relative to base level	L T⁻¹	m yr⁻¹
E	fluvial incision rate; E = K A^m Sⁿ	L T⁻¹	m yr⁻¹
K	bedrock erodibility	L^1−2m T⁻¹	depends on m
A	upstream drainage area	L²	m²
S	channel gradient along the flow path	1	—
m, n	stream-power exponents	1	—
G	deposition coefficient	1	—
Q_s	volumetric sediment flux through a node	L³ T⁻¹	m³ yr⁻¹
q_s	hillslope sediment flux per unit width	L² T⁻¹	m² yr⁻¹
K_d	hillslope diffusivity	L² T⁻¹	m² yr⁻¹
∇z	ground-surface gradient	1	—
S_c	critical gradient the material can hold	1	—
w	channel width	L	m
Q	water discharge	L³ T⁻¹	m³ s⁻¹
ζ	bank migration rate	L T⁻¹	m yr⁻¹
u_b	near-bank excess velocity	L T⁻¹	m s⁻¹
R_c	radius of curvature of a bend	L	m
y	sine of latitude	1	—
P₂(y)	second Legendre polynomial — a function, not a stored variable	1	—
T	mean annual temperature	Θ	°C
P	mean annual precipitation	L T⁻¹	mm yr⁻¹

Fluvial Grand Canyon · NASA

Fluvial incision — stream power

E = K A^{m} S^{n}

E incision rate · K erodibility · A upstream drainage area · S channel gradient · m, n exponents (dimensionless)

Erosion rate scales with upstream drainage area A (a proxy for discharge) and local slope S. This is the term that cuts valleys, and in the shipped bake it does 74.3% of all the work.

Consequence the solver had to handle: on a drainage divide A is one cell by definition, so stream power is ~0 there. Summits cannot erode by this law at all — which is precisely why a hillslope term is required, and why its absence is visible.

The sediment film shows what this law builds. This one shows what it is. Base level falls once, at the right-hand outlet, and then nothing else is done to the landscape. For n = 1 the incision law is a wave equation: the step does not decay where it stands, it travels upstream at celerity c = K A^m. That is a knickpoint — a waterfall, in the field — and it is why a river whose outlet has been lowered rejuvenates from the mouth up rather than everywhere at once. Watch the retreat rate in the readout. It runs fast down the trunk where A is large and stalls as it climbs into the small tributaries, because the celerity carries A^m with it. Nothing scripted would slow down there. No deposition term at all (G = 0), deliberately: this is the bare incision law, so the difference from the sediment film below is exactly the difference between one term and two. Toy domain. Not the shipped bake.

Hillslope Andes · NASA

Threshold hillslopes

q_{s} = \frac{K_{d} \nabla z}{1 - {| \frac{\nabla z}{S_{c}} |}^{2}}

q_s sediment flux per unit width (m² yr⁻¹) · K_d hillslope diffusivity (m² yr⁻¹) · ∇z surface gradient · S_c critical gradient — flux diverges as ∇z approaches S_c

Roering, Kirchner & Dietrich (1999)

Non-linear soil creep: flux rises smoothly at low gradient and diverges as slope approaches the critical angle S_c, which caps hillslopes at a repose-like limit instead of letting them steepen indefinitely.

Measured in the shipped bake: 0 km moved, 0.0%, hillslope faces active 0. The talus term (25.7%) and fluvial term (74.3%) account for all of it. This is reported rather than hidden — the law is implemented and contributed nothing to this particular world, and finding out why is open work.

A scarp the day it was made, and then left alone. Linear diffusion — the denominator deleted — relaxes this into a convex bulge and has no maximum angle at all: feed it a cliff and it will hold a 70° face for as long as you care to watch. The term 1 − (|∇z|/S_c)² sends the flux to infinity as the gradient approaches S_c, so no slope can sit above it for more than an instant and the relaxed profile comes out planar rather than convex — which is what a real threshold hillslope looks like. The checkable claim is in the readout and it is an output: the maximum gradient starts at roughly 13× the threshold and then asymptotes onto it — measured off this build: 1.85× S_c a third of the way through, 1.02× at the end, and it never crosses. It flattens out there rather than continuing to decay because the instant a slope drops below threshold the flux amplification collapses with it, and the channel at the foot keeps cutting down and holding it up. No line in the solver says do not exceed. One honest caveat: that amplification is capped at 12× for numerical stability, which is why the maximum settles a few per cent above S_c rather than exactly on it. Toy domain. Not the shipped bake.

Sediment Lena Delta · NASA

Sediment transport & deposition

\frac{\partial z}{\partial t} = U - K A^{m} S^{n} + \frac{G Q_{s}}{A}

z elevation · t time · U uplift rate · K A^m Sⁿ the incision term E above · G deposition coefficient (dimensionless) · Q_s volumetric sediment flux through the node · A drainage area. Q_s/A has dimensions of a rate because a volume flux over an area is a thickness per unit time.

Davy & Lague (2009) · coefficient range from Yuan et al. (2019)

The transport-limited counterpart to pure incision. G sets how strongly suspended load redeposits: G = 0 is detachment-limited (rivers only cut), G > 0 lets valley floors aggrade, building fans, floodplains and deltas.

Shipped value G = 0.03. The bake logs record the search that landed there — G = 1.0, then 0.1, then 0.03, with the last two bakes both at 0.03 — and at every one of those values 78.2 % of cells received sediment. Coverage therefore could not separate them at all; the value was chosen on how much was laid down, not on whether anything was. Predictions were written before the bake: valley floors must rise, relief p99 must fall slightly, and sediment delivered to the sea must exceed zero — a closed planet that erodes but never exports is mass-conserving only by accident.

The term that is easy to skip is G Q_s/A, and this is what it does. A range uplifts on the left. Flow is routed downhill, drainage area A is summed upslope-first, each cell incises at K A^mSⁿ — and the material it removes is carried downstream and put down again where the slope can no longer hold it, building the wedge in the basin. Red lines are cutting, gold lines are aggrading. The readout tracks eroded against deposited: those two numbers track each other because the transport step conserves volume, which is the one thing a scripted morph between two shapes could not reproduce. Toy domain. Not the shipped bake — the bake ships G = 0.03 and routes multiple-flow; this runs hotter so a fan appears inside ten seconds.

Discharge Amazon · NASA

Channel width — hydraulic geometry

w \propto Q^{0.5} Q \propto A^{0.7 - 0.8} ⟹ w \propto A^{0.35 - 0.4}

w channel width · Q water discharge · A drainage area — the same A as the incision law, which is what lets the two compose

Leopold & Maddock (1953), measured across four orders of magnitude of discharge

At the exponent 0.4, a confluence that doubles drainage area widens the channel by 2^0.4 ≈ 1.32× — a third wider below the junction than above it, which is what a real confluence looks like.

Why the exponent, not the constants: an earlier logarithmic width law widened a doubled river by 4·log₁₀2 ≈ 1.2 m on a 3–14 m half-width. No tuning could fix that, because the shape of the curve was wrong. The reported symptom was exact: rivers now merge instead. the merged river however fails to become larger.

Meanders Mississippi · NASA

Meander migration

ζ \propto u_{b} peak at \frac{R_{c}}{w} \approx 2 - 3

ζ bank migration rate · u_b near-bank excess velocity · R_c radius of curvature of the bend · w channel width, from the width relation above

Howard & Knutson (1984); Ikeda, Parker & Sawai (1981); Hickin & Nanson (1975)

Bend migration is driven by near-bank excess velocity with a weighted memory of upstream curvature, so bends grow, sharpen and eventually cut off.

The counter-intuitive part, and the reason the naive version was wrong: migration rate is not proportional to river size. Hickin & Nanson measured it peaking at an intermediate curvature-to-width ratio and falling for tighter bends.

Plan view, not a profile — this law is about a line, not a surface. The migration rate is driven by the near-bank excess velocity, and the part that matters is that the excess is not the local curvature: it is a weighted memory of the curvature upstream, decaying as exp(−ks). Each bend is steered by the bend above it. That single non-local term is why meanders migrate downstream and lean over instead of growing symmetrically in place. The checkable claim is in the readout: sinuosity climbs from 1.00 to about 3.3, and along the way a neck cutoff fires — two of them in this clip — and each one removes an entire loop in a single step. Watch the sinuosity figure drop when it does. Over geological time that is the whole reason meandering rivers have a sinuosity ceiling at all: length is added continuously and removed in jumps. A scripted animation would grow forever, or stop where it was told. The migration rate also carries Hickin & Nanson’s limiter, which is printed on the card beside this film: rate peaks at an intermediate curvature and falls for tighter bends. Without it the leading bend accelerates into itself and winds up as a spiral no river makes — which is exactly what the first build of this film did. Pale loops are abandoned oxbows. Every bank here is equally erodible, which is why the pattern is so regular; a real valley is not, and neither is the bake’s. Not the shipped bake.

Ice line Malaspina Glacier · NASA

Insolation & the ice line

P_{2} (y) = \frac{1}{2} (3 y^{2} - 1) warmth = \frac{1 - P_{2}}{1.5}

y sine of latitude (0 at the equator, ±1 at the poles) · P₂(y) the second Legendre polynomial — a function of y, not a stored field, and not the precipitation P below. Warmth is normalised so the equator is 1 and the poles are 0.

Second Legendre polynomial — the standard latitudinal insolation shape in EBMs · North (1984)

Warmth follows the second Legendre polynomial in y = cos(colatitude), the classic energy-balance form. The equilibrium line altitude is derived from this climate, not authored.

North's small-ice-cap instability is respected: a cap shrinking past a critical size disappears entirely rather than thinning away smoothly, so polar caps are bistable.

This world's poles are on the X axis (PolarAxis = (1,0,0)) — worth knowing, because assuming Y produced a map whose cold biomes sat at mean |lat| 22.9°. Re-projected about the true axis they sit at 58.6°, and the tropics fall to 0.03%.

Biomes Kakadu · NASA

Biome classification

biome = f (T, P)

T mean annual temperature (°C) · P mean annual precipitation (mm yr⁻¹) — the two axes of the Whittaker plane

Whittaker's temperature × precipitation scheme

Desert, savanna, rainforest, grassland, forest, taiga and tundra are assigned on the Whittaker plane rather than by altitude bands, so a cold-wet cell becomes boreal forest and a cold-dry one becomes tundra.

Freshwater zonation is limnological, not invented: a supralittoral band, an emergent macrophyte belt, then the profundal — because our lakes run 44–68 m deep while rooted plants stop at 3–10 m, so most of a lake's area is genuinely below the photic shelf.

Karst Karst, China · NASA

Cave morphology

branchwork ≫ network ≫ anastomotic ≫ ramiform ≫ spongework

Palmer (1991), on the distribution of solutional cave patterns

Cave families are generated in the proportions Palmer measured in real karst. The note in the source is the useful part: an earlier generator produced spongework, which is the rarest pattern in nature — a plausible-looking result that was statistically almost impossible.

Glacial Shallow-ice approximation

Ice flow & the glacial buzzsaw

q = Γ H^{n + 2} {| \nabla s |}^{n - 1} \nabla s

q ice flux · H ice thickness · s ice SURFACE (bed + H) · n Glen's exponent = 3 · Γ deformation constant

Glen's flow law; abrasion E_g = K_g U_s^l after Humphrey & Raymond (1994), exponent revisited by Herman et al. (2015)

Stream power is E ∝ A^mSⁿ and drainage area is one cell on a divide by definition, so rivers cannot erode a summit. Ice can — it flows from the divides. That is what cuts cirques back into a peak, leaves horns and arêtes between them, and replaces V-notches with U-troughs. Mass balance is keyed to the equilibrium line altitude: b(z) = clamp(γ (z − ELA)), accumulating above it and ablating below.

Note it is the ice SURFACE slope, not the bed slope. That one substitution is why glaciers overdeepen — they can erode below their own outlet — which is the origin of finger and ribbon lakes. Abrasion peaks where ice is thick and moving, mid-glacier near the ELA, so summit height is limited by physics rather than by a hand-tuned clamp. The solver relaxes toward the SIA steady state with a bounded per-face transfer instead of sub-cycling an explicit scheme whose stable timestep collapses as H⁵: correct where the landscape ends up, wrong about how fast it gets there, which is the right trade for a bake.

This is the equation above, integrated, not an animation of one. The block starts as a stream-power landscape — note the V-shaped notch, which is what E = K A^mSⁿ cuts. As the ELA drops, ice accumulates where b = β(s−ELA) is positive, spreads by the shallow-ice flux q = ΓHⁿ⁺²|∇s|ⁿ⁻¹∇s, and cuts the bed at ė = K_gU_s. The U-trough is an output, and the readout measures it: the section index runs 0.5 → 1.0, which is a linear V becoming a flat-floored U. It happens because abrasion acts normal to the bed, so the same cut on a steep wall lowers the surface far more and cuts the wall back — erosion that merely followed ice thickness would deepen the V instead, and in testing it did. Watch the summit figure fall too: that is the buzzsaw. Then the ELA rises again and the ice retreats, which is what uncovers the trough. Toy domain, ~1,500 cells, seconds standing for 10⁵ years. Not the shipped bake, not its resolution, not its constants.

Precipitation Smith & Barstad, 2004

Orographic precipitation

U \frac{d q_{c}}{d s} = S - \frac{q_{c}}{τ_{c}}, P = \frac{q_{s}}{τ_{f}}

q_c cloud water · q_s hydrometeors · τ_c conversion time · τ_f fallout time · S = C_w max(U·∇h, 0) forced uplift

Smith & Barstad (2004), A linear theory of orographic precipitation

Two coupled advection–relaxation equations along the wind. The max(…, 0) in the source term is the rain shadow: air descending a lee slope does not condense, so the lee gets only what survives the fallout timescale. One ridge therefore becomes a rainforest flank and a desert flank — and on this world that matters more than temperature does, because the measured pole-to-equator contrast is only 2.50 K. Precipitation is what carries the map.

Solved in real space, not with an FFT, and the reason is a scar. The published solver is a Fourier transfer function. A per-cube-face FFT cannot see across a cube edge, and this generator has already paid for that once — great-circle arcs of fake mountains, seam ratio 1.74× driven back to 1.02×. The real-space upwind march is identical physics with identical parameters and inherits seam-freedom from the cross-face neighbour topology instead of fighting for it again. Information travels downwind only, so sorting cells along the wind makes a single pass exact.

Delete the two delays and this becomes a cartoon. With τ_c and τ_f removed, precipitation is proportional to uplift: rain exactly on the windward face, none past the crest, perfectly symmetric. The delays mean water is lifted here and falls there, and there is Uτ downwind. The checkable claim, and a cartoon cannot make it: in its last third the film raises the wind from 8 to 22 m/s, and the precipitation maximum crosses the crest. Measured off this build: at 8 m/s the peak sits 0.7 % of the transect UPWIND of the summit; at 22 m/s it is 7.6 % DOWNWIND of it, and the far lee goes from 5.7 % of the peak to 22.8 %. A faster flow carries the drops further before they land. That is why real ranges are wettest just over the crest rather than on the face, and why a rain shadow deepens when the flow is strong. The blue curve is accumulated precipitation, the pale verticals are cloud water aloft. A 1-D transect: there is no cross-wind term here at all.

Coastal Effective fetch · NOAA WEMo

Wave exposure & the coastal ecotone

16 great-circle rays · ≈45 samples per ray, log-spaced · breaker index γ ≈ 0.78 · shoal termination at 2.5 m

Effective fetch / relative exposure index; Burrows' wave-fetch model; NOAA WEMo

Mangrove, tidal flat, salt marsh and estuary are decided by six per-cell tests — tidal frame, flatness, shelter, thermal class, substrate, ponding. Two of those had no field behind them, and they are precisely the two that discriminate: build the biomes first and you gate on a test that always returns true, which is how you get mangroves on cliffs.

A shoreline is a 1-D feature in a 2-D grid — order 0.2–1% of the planet at 19 m cells — and that single fact is what makes the expensive, correct method affordable: real ray-cast fetch on a quarter-million cells instead of a blurred mask on a hundred million.

Every cheap alternative fails in the same place. Distance transform, gaussian blur, mip-pyramid land fraction, diffusion — none can represent DIRECTION, and direction is the entire difference between a lee shore and a windward one. Sixteen rays because eight aliases: a 45° gap either fully counts or vanishes, so bays flicker between sheltered and exposed under tiny coastline changes. Rays march the great circle in 3-D as p cos a + t sin a, which makes cube seams a non-issue by construction rather than by patching.

Reef Photic zone, 2–40 m

Reefs as a constructional landform

no reef above 2 m · dense cover 5–25 m · framework stops at 40 m · shelf break derives at ≈162 m

Hermatypic coral growth; light as the master control

The only landform here that is built by life: coral lays down carbonate and grows toward the light, so the rock is the organism's output rather than the solver's. It writes height and hardness and then gets out of the way — the same shape the volcanic and canyon passes use, and for the same reason.

Capping growth just below sea level produces all three reef types for free: a shore with nothing to enclose gets a fringing reef; a rim that caps while the ground behind stays deeper is a barrier reef with a lagoon; the same rim around a drowned volcanic edifice is an atoll.

Below 2 m there is no reef, and that is the counter-intuitive half. The reef FLAT is the harshest zone on the whole structure — emersion at low tide, thermal and salinity extremes, wave shock — so it is pavement, rubble and algal ridge, never the coral garden. Substrate splits the result in two: on soft bottom the animal builds its own framework (Reef), on rock the rock reads through a veneer (RockyReef). Invert the substrate test, drop the construction, and the same pass yields Seagrass — which in reality covers far more sea floor than coral.

Two rules. Three landforms. The solver does not contain the word “atoll”. Coral accretes at dC/dt = G (1 − depth/D_photic) wherever it is under water and above the photic floor, and may not grow past sea level. And the rate scales with exposure — the fraction of eight compass rays from a cell that reach open ocean without crossing a crest. That second rule was not in the first build, and the film is what said so: light alone fills the entire shallow platform, because a cell behind the rim is lit exactly as well as a cell on it, and the readout printed lagoon depth 0.000. What starves a back-reef is not darkness, it is still water. Nothing else is written down. Then the island subsides. While it stands high the reef hugs the shore — fringing. As the ground sinks the reef keeps pace with sea level but the shoreline retreats behind it, and the water between them deepens: that gap is a lagoon and the reef is now a barrier. When the last of the island goes under, the rim is still there — an atoll. That is Darwin’s 1842 argument, and here it is an output. The readout names the stage, and it is computed from the grid — land cells, and water enclosed between two crests — never from the frame number. Sea level never moves; the ground does, which is the choice the bake makes and the reason the subsidence is printed. Toy domain. Not the shipped bake.

Volcanic Morphometry, one parameter

Volcanic edifices

h / w \approx 0.10, \bar{θ} \approx 12 °

shields 2–10° · stratovolcanoes 25–35° · one viscosity parameter spans the continuum

Volcano morphometry: the shield/strato split is a single number, not two species

An edifice is a constructional landform — lava erupted and piled up — so writing height is the physics, not a shortcut. It is stamped before the erosion solver with a hardness anomaly beneath it, so the solver then cuts radial gullies into the flanks, dissects the soft ash and leaves the resistant core. That dissection is what stops it reading as a smooth cone-shaped bump.

Low-viscosity basalt flows before it piles (shields); viscous gas-rich andesite piles where it lands (stratovolcanoes). One knob spans both — exactly what an exoplanet generator wants.

The caldera costs nothing and pays for itself. The summit carries a depression and the bake already segments lakes out of the depression fill, so a caldera fills itself with a crater lake: no lake code, no special case, just a hole in the right place. Islands are the same feature placed on the shelf — a cone tall enough to breach sea level from shallow water is a volcanic island, which is how real arcs form, and the drowned ones that fail to breach become the platforms atolls grow on.

Two rules, and the stratocone profile is not one of them. (1) each eruption lays a thickness that falls off with distance from the vent; (2) between eruptions the pile relaxes any face steeper than the angle of repose. That is the whole model, and the concave-up flank of a real volcano comes out of it without being written down anywhere. The checkable claim is in the readout: the flank angle climbs and then stops at repose while the summit keeps rising — which is the observation the shape is usually justified by, arriving here as a consequence instead of an assumption. And the caldera is not a dent. It is the roof of an emptied chamber failing: one eruption an order of magnitude larger than the rest withdraws more than the pile can bridge and the summit drops into the volume it just erupted. Nothing is subtracted that was not first put there. Toy domain. Not the shipped bake.

Canyon A ratio, not a shape

Canyons

repose = lerp (θ_{soil}, θ_{rock}, H)

H hardness · incision ∝ A^m concentrates power into a line · wall retreat relaxes to repose

A canyon is not a shape, it is a ratio: incision must beat wall retreat. Every term that decides it was already in the pipeline. Stream power cuts as A^m, so a big river concentrates its power into a narrow line — that is the cutting. Thermal erosion relaxes slopes to an angle interpolated by hardness, so hard rock already stands steeper — that is the wall. Uplift supplies the vertical range — that is the depth.

The load-bearing coincidence: the one field that resists widening is the same field that lets the walls stay vertical, so a single hardness anomaly buys both halves and no new solver is needed. The pass therefore writes a plateau and a resistance and lets the existing river cut it, which comes out sinuous instead of stamped.

Eight of these fourteen cards carry a film, and here is why the other six do not. A film is worth building only where the law describes something that moves, so the absence is a judgement and it is stated rather than left to look like an oversight. Channel width and biome classification are not processes — one is a scaling relation and the other a classifier, and integrating either forward would produce a picture of nothing happening. Insolation and the ice line is a curve, and the thing it drives is already the glacier film’s whole plot. Canyons integrate the same two terms as the fluvial and hillslope films with one parameter changed, so a third film would repeat two you have already watched. Wave exposure and cave morphology are the two honest gaps: both are genuinely animatable, both need a solver of their own — a plan-view fetch cast and a 3-D dissolution network — and neither is written yet.

Quantity	2026-08-12	now	why
World	riverworld	riverworld50	the scene binder's `bakeResource`; every consumer now derives its filenames from it rather than hard-coding a name
Radius	25,000 m	50,000 m	—
Face resolution	1024²	4096²	16× the samples per face; 19.2 m per cell
Height samples	6,291,456	100,663,296	—
Blob	123.8 MB	2.56 GB	moved `Resources/` → `StreamingAssets/` and streamed: Unity's single-object ceiling is 2 GB, so a 4096-face blob could not be a `TextAsset`
Format	PLNT v6	PLNT v7	+ per-cell Sediment and Precip fields, appended after the v6 prefix
Relief scale	1,536.1 m	2,607.1 m	—
Height range	−566 … +2,767 m	−1,424 … +4,367 m	—
Ocean	20.4%	54.5%	the earlier figure was a hypsometric estimate on a unimodal curve; this is a cell count of the two marine classes. 62.2% of the surface is below sea level once beach and reef are included
River splines	1,019	1,918	2,201,433 nodes
Lakes	353	3,303	—
Biome classes	17	35 declared, 33 present	see below
Bare rock	22.94%	3.85%	the largest single correction on this page. A generic 20–40° slope ramp was painting bare cliff onto every soil hillside the erosion solver had built — the angle of repose is 32°, which is 53% of the way up that ramp. Grassland went from 0.66% to 8.31% in the same fix

Quantity	gnomonic	equi-angular	why it matters
Cell area ratio, max/min	5.196	1.4099	—
Cells within ±20% of mean	32.5%	100.0%	more than half the grid used to be finer than the quality bar it was sized against, and paid full memory and full bake time for it
Coarsest cell	2R/N	R·(π/2)/N exactly	the shipped detail is set by the worst cell, and it is now the number the code always claimed
Seam relief bias	1.74	~1.00	the visible one. Erosion computes slope as `dh/spacing` and was fed a constant up to 1.67× wrong toward the seams, so it systematically under-eroded there and inflated seam peaks. Removed at the source rather than compensated
Seam adjacency table	valid	valid, unchanged	the warp is separable and fixes s = ±1 exactly, so every face boundary lands on the same great circle. An exact equal‑area map would not survive this: it is non‑separable, and breaks 4 of the 12 seams by up to 76.5 cells

Class	Province	Cells	of sphere	of land	What it is
DeepSeabed	Ocean floor	43,797,330	43.509%	—	abyssal plain
Shelf	Ocean floor	11,085,693	11.013%	—	continental shelf
Grassland	Grass & steppe	8,366,960	8.312%	19.04%	temperate + dry
Beach	Coast	5,672,439	5.635%	12.91%	marine sand at the waterline
Rock	Rock & alpine	3,874,620	3.849%	8.82%	exposed bedrock
Forest	Forest	3,868,336	3.843%	8.80%	temperate + wet
Desert	Arid	3,847,136	3.822%	8.76%	hot + dry
Savanna	Grass & steppe	3,549,152	3.526%	8.08%	hot + medium
Snow	Ice & tundra	2,364,994	2.349%	5.38%	wind-drifted surface
Dunes	Arid	1,848,202	1.836%	4.21%	erg — the one biome that writes height
Tundra	Ice & tundra	1,733,306	1.722%	3.95%	cold + dry
Rainforest	Forest	1,608,222	1.598%	3.66%	hot + wet
Reef	Reef & meadow	1,472,425	1.463%	—	carbonate framework, 2–40 m
AlpineMeadow	Rock & alpine	1,222,094	1.214%	2.78%	treeline to bare rock
Wetland	Fresh water	1,033,605	1.027%	2.35%	marsh and floodplain
Steppe	Grass & steppe	880,104	0.874%	2.00%	high, flat, dry continental
ConiferForest	Forest	685,984	0.681%	1.56%	cold-temperate + wet
Canyon	Rock & alpine	660,132	0.656%	1.50%	stratified wall rock
SaltMarsh	Coast	649,751	0.645%	1.48%	herbaceous, above mean high water
Scree	Rock & alpine	538,623	0.535%	1.23%	talus the slope shed
Ice	Ice & tundra	426,529	0.424%	0.97%	glacier body, walkable
Taiga	Forest	383,409	0.381%	0.87%	cold + wet boreal
Seagrass	Reef & meadow	191,873	0.191%	—	soft-bottom photic meadow
RockyReef	Reef & meadow	182,471	0.181%	—	coral veneer on rock
Swamp	Fresh water	180,422	0.179%	0.41%	ponded, saturated, vegetated
LakeBed	Fresh water	131,421	0.131%	0.30%	submerged freshwater bed
VolcanicAsh	Rock & alpine	131,241	0.130%	0.30%	tephra flanks the solver gullies
LakeShore	Fresh water	66,618	0.066%	0.15%	emergent macrophyte margin
Basalt	Rock & alpine	66,363	0.066%	0.15%	cone and caldera rim
SaltFlat	Arid	62,675	0.062%	0.14%	evaporite in a closed basin
Moraine	Rock & alpine	50,157	0.050%	0.11%	till at maximum ice extent
TidalFlat	Coast	17,393	0.017%	0.04%	bare intertidal mud and sand
Estuary	Coast	13,616	0.014%	0.03%	drowned low-energy river mouth
Mangrove	Coast	0	0.000%	0.00%	intertidal forest on anoxic mud
Soil	Unclassified	0	0.000%	0.00%	altitude-only fallback — never reached

A planet built from
published geomorphology

The process laws

Fluvial incision — stream power

Threshold hillslopes

Sediment transport & deposition

Channel width — hydraulic geometry

Meander migration

Insolation & the ice line

Biome classification

Cave morphology

Ice flow & the glacial buzzsaw

Orographic precipitation

Wave exposure & the coastal ecotone

Reefs as a constructional landform

Volcanic edifices

Canyons

Erosion budget of the shipped world

What changed since this page was written

The projection was wrong, and two constants had been right all along

Every baked world before this is invalid, and no checksum could tell you

The planet no longer has to fit in RAM

And the bake stopped hoarding

What the world is made of

The climate this world was solved for

Evolution of the bake

Provenance