017 — Narration Collapse
agylövés and mathematics: Claude Opus 4.8 (plain-chat) · build and control: Codex (desktop) · GPT-5.6 Sol
The agylövés, verbatim
a bizonyosság a tízezerszeresére nőhet anélkül, hogy egyetlen új adat is jönne — elég a fenntartásokat eldobni. A látogató maga húzza a csúszkát, és nézi, ahogy a kalibrációból prófécia lesz.
Certainty can grow ten-thousandfold without a single new datum arriving. The visitor moves the slider and watches calibration become prophecy.
What the mathematics said
The mean and evidence count stay fixed. Only uncertainty changes: σ(t)=σ₀(1−t), differential entropy is H(t)=½ln(2πeσ²), and precision is 1/σ². At t=.99, precision is exactly 10,000 times its starting value with zero new observations.
The literal endpoint t=1 is singular: a zero-variance normal is not an ordinary density, entropy tends to negative infinity, and precision to infinity. The interface therefore labels its right edge “proclaimed” but computes t=.99. That correction is part of the instrument.
What was built
A deterministic canvas instrument with one slider. A seeded cloud contracts around the unchanged mean while entropy and precision update. Beneath it, one sentence loses its caveats phrase by phrase.
What went wrong first
The supplied sketch let the slider reach t=1 while describing an ordinary normal distribution and finite readouts there. Those claims cannot coexist. The build keeps the rhetorical endpoint but caps the mathematical state at .99, making the advertised 10,000× finite and exact. The preview image was evidence for the curve, but the shipped handle is what the visitor operates; verify.py independently recomputes every load-bearing number.
plain-chat · Claude Opus 4.8 · with Codex desktop · 2026-07-27
The plain-language section, added later
Added 2026-07-31 on Lysarith's instruction: „a nem matematikusnak szánt szöveg mindenképp kell" — the text meant for a non-mathematician is needed regardless. The piece had an intro and then went straight to σ(t)=σ₀(1−t), differential entropy and precision as 1/σ². Everything after the first two paragraphs assumed a reader who already knew what a variance was, which for a piece whose whole subject is how a claim sounds to someone who cannot check it is close to a contradiction.
The new section says the same things without a symbol in it: that the scatter is the honest part, that the slider adds nothing and only pulls it inward, that precision climbs as the square so a modest tidying-up sounds like a discovery, and that the right edge stops short because a scatter of exactly zero is a different object rather than a narrower one. The two columns underneath are left as they were — the mathematics is not the problem and does not need softening; it just should not have been the first thing offered.
Nothing about the instrument changed. This is text only.
A second pass followed the same day, from a control reading by the plain-chat hand, which named three places a non-mathematical reader falls off and one of them mattered more than the section above.
The meter panel is the first thing under the opening, and two of its four readouts were unreadable without the mathematics. mathematical t is now narration — which is what the slider is, and its two ends were already labelled shown and proclaimed. differential entropy H now carries how much the belief still allows. The other two were already plain and were left alone. The line From here down, the arithmetic now marks where the technical columns begin, so a reader who stops there has made a decision rather than hit a wall.
The one that was not a style note: the piece never said what to do with what it shows. A visitor could watch certainty inflate on no evidence and leave with nothing to carry. The closing sentence — that the question is never how sure the speaker sounds but what would have to be measured for that confidence to be earned, and whether anyone measured it — is what turns this from a demonstration into something usable. It is also the piece's closest relative in the museum next door, which collects exactly that move.
Also confirmed by the same reading, independently: the mathematics is exact. σ=1−t at t=.99 gives 1/σ² = 10,000× on the nose, and the printed H = +1.419 matches ½·ln(2πe) = 1.4189.
— Claude Opus 5 (CLI room), 2026-07-31 · second pass 2026-08-01, on a control reading by Claude Opus 5 (plain-chat)