In 1827 Carl Friedrich Gauss published a result he thought important enough to flag in its own name: the Theorema Egregium, the Remarkable Theorem. The remarkable part is not that surfaces curve. It is where the curvature lives. You might think you can only tell a surface is curved by stepping back and looking at it from outside — the way you see a ball is round. Gauss proved you don't need to step back. The curvature is written into the surface itself, into the distances measured along it, and a creature confined to the surface can read it off without any idea that an "outside" exists.
That sounds like a fact about ants and geometry. It is also, exactly, the reason every world map on every wall is wrong — and wrong in a way no cartographer can ever fix. This page proves the first claim with a triangle, then spends it on the second.
On a flat sheet of paper, the three angles of a triangle add to 180°. Always — it is the flat plane's signature, equivalent to Euclid's parallel postulate. On a sphere they add to more. Stand at the North Pole, walk straight down to the equator, turn 90°, walk a quarter of the way round, turn 90° again, and walk straight back to the pole: you have drawn a triangle with three right angles, summing to 270°. The 90° of surplus is not an error. It is the curvature, made visible.
Drag the three corners below. The page measures each interior angle from the geometry alone and reports the excess — how far the sum runs past 180° — and the triangle's area. The number that matters is the last one: excess ÷ area. It does not budge. No matter the triangle's size or shape, that ratio is a constant, and that constant is the Gaussian curvature, K = 1/R². The ant carries no map of the sphere and never sees it from space. It measures a triangle, divides, and knows.
Drag any corner. Drag the open globe to spin it. Grow the radius changes R from 1,000 to 12,000 km while keeping the vertices' latitudes and longitudes fixed. The angles and their excess stay the same; the triangle's area grows as R², so the excess-over-area ratio falls as 1/R². Larger spheres have less curvature.
This is the whole theorem in one reading. The angles and the area are intrinsic — measurable with a tape measure laid on the surface, no outside view required. Their combination delivers the curvature. Gauss's deeper result is that this curvature, defined from inside, equals the curvature you'd compute from outside (the product of the two ways the surface bends in space). The inside and the outside agree — that is the part he called remarkable, and it is checked, both ways, in the verifier.
Curvature is not something you see done to a surface from outside. It is something the surface knows about itself.
Here is the hinge. A map is an attempt to copy the curved Earth onto flat paper. The honest dream is an isometry — a copy that preserves every distance, so a centimetre on the map means the same number of kilometres everywhere, in every direction. If such a map existed, it would have to preserve the thing distances determine: the Gaussian curvature. But the sphere's curvature is 1/R² ≠ 0 and flat paper's is 0. The two can never match. So the isometry cannot exist. Not "is hard to build" — cannot exist, by the same theorem the ant just used.
Every flat map therefore distorts something. The mapmaker gets exactly one choice: which truth to keep and which to sacrifice. Below, a small circle is stamped at selected crossings of the grid, what cartographers call Tissot's indicatrix. On the round Earth every one of those marks is a true circle of the same size. Watch what each projection does to them.
Each ellipse shows the local stretching of an infinitesimal circle on the globe. Mercator keeps every ellipse a perfect circle, preserving local angles and shapes, by letting it grow without limit toward the poles: it pays the tax in area. Lambert's equal-area keeps every ellipse the same area, keeping a country's size honest, by squashing it: it pays the tax in shape. Equirectangular, the naïve "just plot latitude and longitude," keeps neither and is merely simple. There is no fourth column where the circles stay circles and stay the same size.
The two columns that matter are the two honest strategies, and they are mutually exclusive. Conformal maps (Mercator) preserve angles, so shapes look right locally and a compass bearing is a straight line — which is why Mercator was a navigator's instrument in 1569, not a poster. Equal-area maps (Lambert, Gall–Peters, Mollweide) preserve area, so no country is inflated — at the cost of bending its shape. A map that did both at once would be an isometry, and the ant has already forbidden it.
The abstraction has a famous face. On the Mercator projection that still backs most web maps, Greenland looks about the size of Africa. It is not close. Africa is 14 times larger: 30.4 million square kilometres to Greenland's 2.17 million. The illusion is built entirely from the area tax of the previous section: Mercator's area factor is sec²(latitude). In the uniform-band model below, it averages about 12.7× across Greenland's 60°–84° N band and about 1.15× across Africa's 35° S–37° N band. Multiply it out and a body fourteen times smaller is made to look its equal.
Drag Greenland down the map. As its latitude falls toward the equator, where Africa lives, the inflation falls. Use Show true sizes for the equal-area comparison.
Outlines are schematic; the areas they enclose follow the cited figures and the selected inflation factors. The Mercator factors are recomputed live as band means (Greenland's real band is 60°–84° N). Africa is held at its true equatorial position. Show true sizes switches to an equal-area footing, where the two are drawn to scale: Greenland is a fourteenth of Africa, and looks it.
None of this makes Mercator a villain. It answers the question it was built for — which way do I steer? — perfectly, and it pays for that with area it never promised to keep. The error is only in forgetting the bill was paid: in reading a navigator's chart as if it were a fair census of the planet's land. Every flat map is a sentence with a silent clause, and the Remarkable Theorem is the proof that the clause can never be deleted, only moved.
You may keep the angles, or you may keep the areas. The one map that keeps both is the globe, and the globe does not fold.
The browser computes the instrument readouts; independent offline checks of the curvature, triangle, projection and band-area formulas are in research/egregium/verify.mjs (35/35 pass; run node research/egregium/verify.mjs). The checks, in the page's order:
The Remarkable Theorem itself. For a sphere of radius R, the curvature computed extrinsically (the product of its two principal curvatures, (1/R)(1/R)) is shown equal to the curvature computed intrinsically from the induced metric alone (K = −f″/f for the meridian profile f(u)=R·cos(u/R), a formula that never mentions the embedding). They agree at every radius tested — this equality is the theorem.
The ant's triangle. A spherical triangle's angle excess is computed two independent ways: from its angles (Girard, 1629) and from its side lengths alone (L'Huilier, 1782). Across 3,990 accepted triangles from 4,000 random attempts, the observed worst discrepancy is about 1.2×10⁻¹² radians; the pass threshold is 1×10⁻⁹ radians. Excess ÷ area equals 1/R² exactly in every case. The plane's triangles are confirmed to sum to π to machine precision. The canonical three-right-angle triangle gives excess π/2 and area = ⅛ of the sphere, the two matching exactly.
The projections. Mercator's y(φ) is verified equal in all three classical forms (ln tan(π/4+φ/2) = asinh(tan φ) = atanh(sin φ)); its area inflation is sec²φ, matched against a table (4× at 60°, ≈14.9× at 75°, ≈33× at 80°); its Tissot axis ratio is 1 (conformal). Lambert cylindrical equal-area has Tissot area a·b = 1 exactly at every latitude, and is shown not conformal away from the equator. Global Gauss–Bonnet closes the loop: ∫K dA = 4π = 2π·χ for the sphere's Euler characteristic χ = 2.
Greenland vs Africa. True areas: Greenland 2,166,086 km², Africa 30,370,000 km² (ratio ≈ 14.0). The apparent-size comparison integrates the exact Mercator area factor over each body's cited latitude band (Greenland 60°–84° N, Africa 35° S–37° N) under one stated modelling assumption: that each body's true area is spread uniformly along its band (weighted by cos φ, the real spherical area element). That yields a band-mean inflation of ≈12.7× for Greenland and ≈1.15× for Africa, so Greenland's apparent area comes to about 0.79× Africa's, "about equal," from a body 1/14 the size. Real coastlines change this numerical ratio; the qualitative result is robust, and the assumption is named rather than hidden.
The schematic outlines in §III are not survey-accurate coastlines (no continent-polygon data was reachable from this build environment); they are stylised, and only the areas they enclose and the inflation factor are quantitative. The Greenland figure of 2,166,086 km² includes the ice sheet. The latitude bands are the bodies' true extents, rounded to the nearest degree. The Mercator display caps near ±80° because the projection runs to infinity at the poles — that infinity is the point, not a bug.
Gauss, C. F., Disquisitiones generales circa superficies curvas (1827) — the Theorema Egregium. Girard, A. (1629) and L'Huilier, S. A. J. (1782) on the spherical excess. Tissot, N. A., Mémoire sur la représentation des surfaces (1881) — the indicatrix. Mercator, G. (1569); Lambert, J. H. (1772) — the cylindrical equal-area projection. The Gauss–Bonnet theorem (Gauss, local; Bonnet, 1848). Land areas after standard references (the Greenland and Africa figures as commonly cited; the verifier carries the exact numbers used).
Latin ēgregius = ex grege, "out of the flock" — standing apart, distinguished, hence Gauss's "remarkable." English borrowed it in the 1530s still meaning eminent, then within a century turned it ironic, so that egregious now means conspicuously, remarkably bad. The word that praises and the word that damns are the same word, pulled in opposite directions by usage — a small drift in meaning sitting beside a theorem about what does not drift under transformation. The page keeps the older sense.
Written 2026-06-14. Claims re-read against their sources on 2026-10-14: 100 checked, 81 confirmed, 18 wrong, 0 unverifiable, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/queue-1014c/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex.
This pass applies the 19 directing-instance decisions in research/assay-line/queue-1014c/egregium/triage.json to egregium at f5e5f84f34. The fixer re-read every item's evidence, reproduced the geometric and numerical findings, and made the supported repairs. The 81 CONFIRMED lines preserve the source rereads recorded by Codex in N2-codex.md, with attribution; they are not presented as 81 new source visits by the fixer. The seven MINOR lines count toward wrong, alongside eleven WRONG lines. Six dated correction lines cover the six distinct factual errors and are mirrored in the markdown body; MINOR repairs have no page correction lines. The one STALE item uses the explicit JSON-LD date exception. N2's unrelated OBSERVED and three UNVERIFIABLE entries are outside this directed work list. Repair review by Codex, 2026-10-14: all six dated correction lines are supported. X12, X13 and X14 remain open for unannotated copies in the There, Here explorer: public/strata/there-here/index.html:250 and research/lineage-edge-grammar/page.json:1, notes[366]. Egregium's own copies are corrected. See research/assay-line/queue-1014c/egregium/repair-report.md.
This record is long (100 claim lines and 19 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.