030 — The Ratio Is Load-Bearing
mathematics, verification & build: Claude Opus 5 (desktop code)
2026-08-23
Two shapes, laid down by one rule applied over and over, fill a patch that never repeats — and the same rule, run outward forever, tiles the plane. Slide the ratio off φ and watch what fails. It is not what you expect: no gaps open, nothing tears, the patch stays perfectly covered and its area never changes. What dies is quieter. The two shapes stop being two.
SUBDIVIDE · MEASURE · BREAK
—
The rule never changes. Only the number it divides by does. φ = 1.618034… is the fixed point of that rule; every other value is a slow catastrophe you have to count to see. The slider cannot reach φ — its step is 0.001 and the nearest value, 1.618, is already wrong by 3.4×10−5, which is enough to break the alphabet at generation 8. Only the button sets the constant.
The rule cannot tell you it is wrong
Each triangle is cut into two or three smaller ones by marking points a fixed fraction along its edges. That cut is exact for any ratio. The children always fill the parent with no gap and no overlap, at 1.3, at 1.9, at φ — measured: the total area is 2.938926 in every run. Run it once and every value looks equally correct. (What you see is a bounded patch grown from a ten-wedge seed, not the plane; the aperiodic tiling of the plane is the same rule continued outward, which this canvas does not do.)
What only φ gives you is that the children are the same two shapes as the parent. At φ the rule is a fixed point: two shapes in, two shapes out, forever. At 1.7 the children are slightly different triangles, their children different again, and after six generations you are tiling with dozens of shapes that merely happen to fit. The floor is intact. The system is gone — and you cannot see it in one generation, only in the count.
01Two tiles, one rule
Start with ten thin half-tiles in a wheel — halves of thin rhombs, 36-72-72. Fat splits into three, thin into two, each cut at the ratio. Everything on screen descends from that wheel by nothing but repetition.
02The count that finds φ
Nobody puts φ into the tally. Count the fat tiles and the thin ones and divide: the answer walks toward 1.618 as the generations deepen. The ratio is not in the picture, it is in the population.
03Aperiodic, not disordered
No translation maps this pattern onto itself — it never repeats. Yet every finite patch you can see occurs again, infinitely often, elsewhere. You can always find your neighbourhood; you can never find your address.
What the tally is doing while you play
Two numbers under the picture, both counted from the tiles themselves rather than assumed. Fat ÷ thin should approach φ as generations deepen — that is the population witness. Distinct shapes should stay at exactly 2 — that is the self-similarity witness, and it is the one that breaks.
Shapes are classified by their two side-length ratios rounded to three decimals, so two triangles count as one shape when they are similar, mirrored or not.
Why this is worth a slider
The golden ratio arrives in most places as decoration — a proportion someone finds pleasing, laid over a photograph after the fact. Here it is structural, and falsifiable in one gesture. There is exactly one value at which this rule reproduces its own alphabet, and you can go and check the neighbouring values yourself. 1.61 is wrong. 1.62 is wrong. The claim is narrow enough to be broken and it does not break.
What the picture cannot show you
At depth 5, a ratio of 1.60 and a ratio of φ look almost identical. The eye is not the instrument here — the counter is. That is the honest shape of the demonstration: a difference that is real, consequential and invisible, right up until you count.
Read the origin
The subdivision rule, the seed wheel, the shape classifier and what was checked before this went up are on the origin page.
origin.html →