014 — The Smell Is the Only Thing That Gets Out

agylövés: Lysarith · mathematics & execution: Claude (CLI) · 2026-07-26

Nothing escapes a black hole. It still smells. A turkey turns in one vortex, Hawking waves from inside the other, and the only thing crossing between them is a thread of aroma — which is exactly the awkward part, because the hole is not supposed to have a scent at all. Page worked out what that scent carries. Sniff once and you learn nothing: the first quantum out is thermal to twelve decimal places. The information is not inside any whiff. It is in how the whiffs agree with each other, and it stays unreadable until you have kept more than half of them.

LIVE RIG · press EVAPORATE

emitted · of0 · 40

entropy of radiation0.0000 bits

information recovered0.000000 bits

this whiff carried— bits

hole size · N qubits 40 emitted 0

The puffs are drawn identically on purpose — that is the physics, not a shortcut. The panel on the right is an illustration of one measured number: the fraction of cells that have settled is exactly the recovered bits over N, taken from Page's formula. It is not a decoding simulation, and nothing here claims you could read a real message this way.

A painted illustration: a turkey roasts on a spit inside a glowing orange vortex on the left, Stephen Hawking waves from inside a blue vortex on the right, and between them two cats sit in the same posture — one solid black with a bell on its red collar, one translucent and made of stars. A single thread of aroma runs from the turkey, through the cats, to Hawking. Below, a funnel falls through a curved grid. — The picture came first, and the piece was built to it. Lysarith wrote the agylövés and attached this: Hawking waving from the other side, a cat that either sits on the quantum horizon or does not, and the smell of a turkey roasting in a vortex. Three things here are not in the sentence — the cat is drawn *twice*, in one posture, both branches at once; the solid one wears a bell nobody asked for; and the aroma is a single unbroken thread, the only thing that crosses between the two vortices. That last detail is the physics, and the rig above is the same picture, computed.

One whiff is weather. The whole plume is a message.

What the maths said

Throw N qubits into a hole and let it emit them one at a time. After k have come out, the radiation and the remainder share a random pure state, and Page's theorem gives the average entanglement entropy of the part you are holding. Before the halfway point that entropy sits at k bits — maximally mixed, indistinguishable from noise — and it falls short of maximal by less than 2^2k−N nats. For a forty-qubit hole the first emitted qubit carries 1.97 × 10⁻¹² bits. Then the curve turns at the Page time, exactly halfway, where it is short of maximal by precisely half a nat, and comes back down. It is its own mirror: S(k) = S(N−k), which is the statement that the whole thing never stopped being pure. At the end the ledger closes — forty bits in, forty bits out.

⟨S⟩ = Σ_j=n+1^mn 1/j − (m−1)/2n
m = 2^k (radiation), n = 2^N−k (hole), m ≤ n
Page 1993 · proved Foong & Kanno 1994

What went wrong first

The first draft asserted that past the Page time each further qubit hands back two bits instead of one, and checked it two qubits after the turn. It failed: 1.9662. The turnover is not a kink — it is rounded, and the asymptotic rate needs a few more qubits to arrive. The instinct was to widen the window until the test went green. Instead the width became the measurement, and it turned out not to be a property of the curve at all: the shortfall decays like 2^2k−N, so demanding the rate to within ε costs log₂(1/ε)/2 further qubits. Predicted 5.0 / 10.0 / 14.9 / 19.9 for ε = 10⁻³ / 10⁻⁶ / 10⁻⁹ / 10⁻¹²; measured 5 / 10 / 15 / 20. The kink is exponentially sharp. Every extra qubit you wait buys two more bits of proof that the rate is exactly two.

Verified three ways, and the third one does not trust the other two: exact harmonic summation, an Euler–Maclaurin expansion (they agree to 8.9 × 10⁻¹⁶ nats), and 400 sampled Haar-random states per point, which reproduce the formula with a worst z-score of 2.21 across eleven points. The first two are one theorem wearing two faces. Only the third actually looked.

The first build drew the curve and nothing else. Lysarith's verdict was one sentence — so we drew a mountain in two colours — and it was right: the shape of the graph is not the idea. The rig was rebuilt around what the picture already knew.

origin ↗