Five layers of this place each prove something is impossible — no faithful flat map of the Earth, no walk crossing the seven bridges once, no sliding the puzzle into half its arrangements, no untying the trefoil. A sixth proves the opposite: that a thing must exist. They read like four locks and a key. They are the same proof.
An ant that never leaves a surface can prove the surface is curved by measuring one triangle: on a sphere the three angles sum to more than 180°, and the excess, divided by the triangle's area, is the curvature K = 1/R² — the same number for every triangle, large or small. Gauss's Theorema Egregium (1827): no distance-preserving bend can change K. The plane has K = 0; the sphere has K > 0; so no flat map of the Earth keeps all distances. Slide the triangle — watch the angle-sum and area both move, and K hold dead still.
The plane's triangle always sums to exactly 180° (K = 0) — flattenable. The sphere's never does (K = 1/R²) — and that single non-zero number is the whole reason every world map lies.
research/egregium/verify.mjs · 35/35 — angle-excess/area = 1/R² by Girard and L'Huilier; Gauss–Bonnet ∮K dA = 2π·χ, χ(sphere)=2.
Königsberg's seven bridges, four landmasses. A walk that passes through a landmass arrives on one bridge and leaves on another — bridges at any in-between stop come in pairs, so its degree must be even. Only the start and end may be odd. Königsberg has four odd vertices (degrees 5, 3, 3, 3) — two too many. No walk. Toggle a bridge and watch the odd-count change; the walk turns possible the instant it reads 0 or 2, and not one moment before.
Click a bridge on the map to remove or restore it.
Σ degrees = 2 × (bridges), always even — so the odd-count is always even, never 1 or 3. Euler answered this in 1736 and invented graph theory doing it.
research/what-the-bridges-knew/verify.mjs · 35/35 — Hierholzer constructs the walk where one exists; brute force over all 20,160 ordered attempts confirms Königsberg has none.
Sam Loyd offered $1000 to slide the tiles home from the 14–15 swap. Nobody collected: it is impossible. Every slide trades the gap with a neighbour — a single transposition — so it flips the arrangement's sign and flips the gap's distance-parity, together. Their combination J = sign × (−1)gap taxicab never changes. Solved boards have J = +1; exactly the J = +1 boards are reachable — exactly half of them. Slide all you like below: J stays locked, and the verdict only goes green when J = +1.
Tap a tile next to the gap to slide it.
Loyd's swap is one transposition from solved, so J = −1 — and no sequence of slides can ever flip a bit that every slide leaves alone. The smallest rearrangement you can do with the gap home is a three-tile cycle. That is why "swap just two pieces" grid tricks are always a swindle.
research/fifteen-puzzle/verify.mjs · 23/23 — full BFS of the 3×3 puzzle finds exactly 181,440 = 9!/2 reachable boards, and J = +1 is shown necessary and sufficient over all 362,880.
Is a tangled loop knotted, or a circle in disguise? Wiggling can only ever fail to untie it. So colour the strands with three crayons under one rule — at every crossing, all three the same or all three different — and count the legal colourings. The plain loop scores 3; the trefoil scores 9. No wiggle (no Reidemeister move) can change that count, so 9 ≠ 3 is a complete proof the trefoil is genuinely knotted — the first knot ever proven non-trivial. Try to colour each below.
Click an arc of the knot to cycle its crayon. A crossing turns green when its three strands are all-same or all-different.
The count is computed live over every possible colouring — 3 for the loop, 9 for the trefoil — not just the one you build. The honest edge: the figure-eight knot also scores 3, fooling three crayons; it takes five to convict it. No single invariant is the whole truth.
research/knots/verify.mjs · 16/16 — Fox colourings and the Jones polynomial (via the Kauffman bracket); this page ports that exact colouring engine.
Now reverse the polarity. At this instant two opposite points on Earth have the same temperature and the same pressure — forced, not lucky (Borsuk–Ulam, 1933). The engine is the very same parity. Walk a ring of weather stations and compare each to its antipode: the difference flips sign across the half-turn, so an odd function — and an odd function must cross zero. Below is its discrete heart (Tucker's lemma): an even ring of ± beads, antipodes forced opposite. Try to arrange the beads so no + ever touches a −. You can't. The escape you cannot find is the match that must exist.
Click a bead to flip it (+ ↔ −). Its antipode flips with it.
Going once around the ring the sign must change an odd number of times (the antipodal rule forces it), and every sign change is a + next to a −. So there is always at least one — no escape. Forbidding the escape and guaranteeing the coincidence are one theorem, read two ways.
research/borsuk-ulam/verify.mjs · 15/15 — Tucker's no-escape verified exhaustively over all antipodal labellings, rings to 16; the located antipodal twin driven to machine residual.
Each of these layers ends on its own surprise and never names the method it shares with the others. The method is the whole of it: to prove a thing cannot be done, find a quantity the permitted moves leave untouched, and show the thing would have to move it. Differential geometry, graph theory, group theory, knot theory, topology — the same sentence, five times.
But honesty is the only thing this place sells, so be exact about what "the same" means here. The five invariants are not the same number. Curvature is a real number; the bridge-count and the puzzle's J and the antipodal engine are parities (mod 2); the colouring is an integer count (solutions of a mod-3 system). What is identical is not the value but the move — and, notably, three of the five reduce to the cheapest invariant in mathematics, a single parity bit. A real observation, not a mystical one. There is even a thread of Euler's name running through it (the odd-degree count he invented for the bridges; the Euler characteristic χ that Gauss–Bonnet ties to curvature) — two distinct theorems sharing a man, named as coincidence, not connection.
Two nearby strata sharpen the lesson rather than join the spine. You Can't Hear the Shape of a Drum is the inverse cautionary tale — an invariant (the spectrum) that is preserved but under-determines, failing to pin the shape it describes. And The Invariant of Relabeling runs the identical move in cryptanalysis: a quantity no relabelling of the alphabet can touch breaks a cipher called unbreakable for three centuries. Same weapon; different quarry.