Five layers of this place end on a number you cannot write down. They do not mean the same thing by it.
Read five of this archive's layers to their last line and each hands you a quantity it cannot quite give you. Incommensurable ends on √2 — a length Euclid proved is not any ratio of whole numbers. The Most Irrational Number ends on the golden ratio, the number that resists fractions harder than any other. How Big Is the Mandelbrot Set? ends on an area no one has ever proven — the most famous shape in modern mathematics, and nobody knows how much of the plane it covers. A Sextillion Ways Home ends on a whole number, the count of five-bell ringings, that is too large for any machine to total exactly. And You Already Know the Rest ends on the entropy of English — a figure everyone reports as a range and no one as a number.
Each says, in its own register: here is a quantity, and you cannot write it down exactly. It is tempting to file all five under one wistful heading — some numbers are beyond us — and stop. That is the mistake this portal exists to correct. "No exact value" is not one condition. It is three, and they are different in kind, and two plain yes/no questions tell them apart:
Q1 — Does an exact value even exist? Q2 — If it does, can we pin it, or name it with finite information?
Answer those two and the five layers fall into three bins — and, the part none of them says alone, the venue prints a differently shaped object for each: a quoted precision, a bracket, or a band. The shape is the honesty. Walk the three below; each recomputes itself in front of you.
Start where the answer is, in a sense, fully known. √2 has an exact value — a single, definite point on the line, the length of a unit square's diagonal. What it has not got is a finite name as a fraction or a decimal: Euclid's proof shows no ratio p/q ever equals it, and the digits never repeat. Yet it is not beyond us at all. A finite rule — its continued fraction — names it completely, and from that rule you can pour out as many exact digits as you like. Pick a number and watch.
| convergent p/q | value | |x − p/q| | |x − p/q|·q² |
|---|
The right-hand column is the deep one. For every irrational, the best fractions hug it about as 1/q² close — but the constant in front is the number's signature, and it is the smallest possible, 1/√5 ≈ 0.447, exactly for the golden ratio (Hurwitz, 1891). That is what "most irrational" means: φ is the number rationals can approach least well, while √2, with constant 1/(2√2) ≈ 0.354, lets them get measurably closer. Either way the lesson is the same. This kind of "unwritable" is solved. The value exists, a finite rule names it, and we reach any precision we ask for. The only thing that fails is the demand that the name be a ratio — and the venue answers by quoting the digits to a stated precision and proving the irrationality. The number resists a representation, not our knowledge.
Now a quantity that is genuinely out past us. The Mandelbrot set has an area — it is a bounded region of the plane, so a definite real number is its size. We just do not know what that number is. Two pieces of it have areas known in closed form — the main cardioid is exactly 3π/8, the period-2 disk exactly π/16 — and those two alone are already most of the whole. Everything past them is measured, not proven. The best rigorous bracket looks like this:
The exact pieces are recomputed above to fifteen places; the bounds are the published record. A rigorous method — the Gronwall area theorem — has been pushed to five million terms and still cannot bring the ceiling below 1.6829, while the best census of 88 trillion points puts the truth near 1.5066. The bracket is real, and it refuses to close. The five-bell ringing count a(5) is the same wall wearing whole-number clothes: a definite integer near 1.11 × 10²¹, estimated by validated sampling, with the exact value sitting just past the memory of any machine we can build. Here the venue cannot quote a value, so it quotes a bracket or an estimate with its error, and names the gap aloud as an open prize. The value exists; we simply have not reached it. Unlike √2, no finite rule we know writes it down.
The third pole is the strangest, and the one the other four layers only gesture at. Ask "how many bits of information is a letter of English?" and the honest answer is that there is no single number to find — not because we are too weak to compute it, but because the question is not fully posed until you fix conventions the world does not fix for you. Watch the simplest version of the quantity move under choices that are all defensible. Below is one fixed passage; nothing about the text changes. Only the definition does.
This is a different animal from the first two. There is no exact value being withheld — there is no exact value at all, because "the entropy of English" is not a constant of nature. It drifts with the alphabet, the corpus, the register, the century, and what you are willing to call a symbol. The venue's only honest move is to report a band and say, in the open, that no single value exists. Mistaking this for the second pole — treating "we cannot pin it" as if it meant "there is a fact we have not reached" — is the precise error the portal is built to catch.
Lay the five layers on the two yes/no questions and the structure is exact. Pick one and read it down.
Read the dial and the three poles separate cleanly. √2 and φ answer yes, yes — the value exists and a finite rule names it — so their wall is the soft one: representational, already solved, defeated by writing the rule instead of the ratio. The Mandelbrot area and a(5) answer yes, no — the value exists but no rule or machine pins it — so the venue brackets it and names the gap. The entropy of English answers no at the first question, and the second never arises: there is nothing to pin, only a band that is the world's own softness, not a defect of the measurement.
And the one line none of the five member layers says: the venue prints a differently shaped object for each pole — a quoted precision, a bracket, a band — and which shape is honest is decided entirely by those two questions. The deepest of the two is the first. To confuse "we have not reached the value" (the Mandelbrot area) with "there is no single value" (the entropy of English) is to mistake the limits of our knowing for the limits of the thing. The companion portal The Limits of Knowing sorts the first kind further — among answers that do exist, which walls are merely tall (resource) and which have no top (computability, proof); its R(5,5) is exactly this portal's epistemic pole for an integer. This portal adds the pole that one cannot reach: the quantity that was never a fixed number to begin with.
Recomputed in your browser and offline (29/29 machine checks pass — research/the-number-that-wont-resolve/verify.mjs):
— Pole 1 is produced, not quoted. The continued fractions are generated by the standard surd algorithm (√2 → [1; 2,2,2,…]; φ → [1; 1,1,1,…]); the convergents are built from the recurrence and shown to be the Pell solutions (p² − 2q² = ±1) and the Fibonacci ratios respectively; √2's digits come from an exact integer square root of 2·10²ⁿ, not a floating call, and match the reference constant to 50 places. The worst-approximability constants are read off the convergent tails: |φ − p/q|·q² → 1/√5 = 0.4472 and |√2 − p/q|·q² → 1/(2√2) = 0.3536, with φ the larger — Hurwitz's theorem, that √5 is the floor, made visible.
— Pole 2 mixes the computed and the cited, and marks which is which. The two exact pieces are recomputed live — main cardioid 3π/8 = 1.17809724…, period-2 disk π/16 = 0.19634954…, together 1.37445, which is 91.2% of the estimate. The rigorous bounds and the census estimate are cited from the member stratum and its primary sources — Hill's lower 1.506303622, the pixel-census estimate 1.5065918849, the Gronwall upper 1.6829 (5,000,000 terms) — and the portal asserts only the ordering: the bracket strictly contains the estimate and its width (0.3085, about a fifth of the value) is positive, so the bracket does not close. a(4)=44 (the last exactly-known extent count) and a(5)≈1.11×10²¹ are cited from The Extent / A Sextillion Ways Home.
— Pole 3 is the portal's own demonstration. The first-order entropy F₁ is computed live over the displayed passage (Jane Austen, Pride and Prejudice, 1813, public domain) under five honest conventions; the same text yields five different answers — letters-only 4.12, letters+space 4.02, case-sensitive 4.25, letters+punctuation 4.14, raw characters 4.24 bits — a spread of 0.23 bit from convention alone, every pair distinct, each F₁ ≤ F₀ = log₂(alphabet). F₀ itself is a choice: log₂26 = 4.700, log₂27 = 4.755, log₂52 = 5.700.
Cited, not recomputed (and marked as such): the true entropy of English as a band 0.6–1.3 bits [Shannon, Prediction and Entropy of Printed English, 1951; Cover & King, 1978] — established and reproduced in the member stratum You Already Know the Rest; Euclid's irrationality of √2 and Hurwitz's √5 theorem [Hurwitz, 1891]; the Mandelbrot area history [Hill 1997; Ewing & Schober; Bittner et al.], earned in full in How Big Is the Mandelbrot Set?.
The honesty line. The portal asserts exactly one thing its five member layers do not: that their shared closing note — a number you cannot write down — resolves into three distinct conditions, sorted by two yes/no questions, each answered by the venue with a differently shaped object. Everything numerical under that claim is recomputed above or cited to the layer that earns it; the one load-bearing distinction, between unreached and nonexistent values, is the portal's own, and it is the line the whole place is built to hold.