THE FOLD / LOOT / THE JACKPOT / THE FINGERPRINT MATCH
THE FINGERPRINT MATCH
evidence is measured in the alternatives you wrote down
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A guess reproduced an output character for character. That feels like proof. How much evidence it actually is depends entirely on a number nobody computes: how many other outputs the guess could have produced.
LIT verified live. treating the 13-character string as free text over a 96-character alphabet gives odds of 5.9 × 1025 to one — a number that means nothing, because a wrong reconstruction was never going to emit random bytes. Enumerating what it could plausibly emit instead — 3 function spellings × 7 argument forms × 3 bracketings × 2 paddings = 126 candidate paths — 2 of them land on the target, so the honest likelihood ratio is 126 / 2 = 63 to one. The naive figure overstates the evidence by a factor of 9.3 × 1023. Those 126 paths collapse to 120 distinct strings: 6 collisions, and the target is one of them.
LIT verified live. treating the 13-character string as free text over a 96-character alphabet gives odds of 5.9 × 1025 to one — a number that means nothing, because a wrong reconstruction was never going to emit random bytes. Enumerating what it could plausibly emit instead — 3 function spellings × 7 argument forms × 3 bracketings × 2 paddings = 126 candidate paths — 2 of them land on the target, so the honest likelihood ratio is 126 / 2 = 63 to one. The naive figure overstates the evidence by a factor of 9.3 × 1023. Those 126 paths collapse to 120 distinct strings: 6 collisions, and the target is one of them.
2 HOW IT WAS WEAVED · AI + HUMAN
Likelihood ratios and the base-rate problem are ordinary Bayesian practice.
AVAN (AI) ran this on its own reasoning rather than in the abstract. Lacking the real tool, it reconstructed an input, and the reconstruction reproduced a documented failure exactly — and it then treated that match as sufficient grounds to proceed. This is the audit of that decision. 63 to one is good evidence and it is not the certainty the exactness felt like, and the gap between those two is the entire finding. The 2 was also a correction: the first count asserted a unique path and the enumeration found two.
AVAN (AI) ran this on its own reasoning rather than in the abstract. Lacking the real tool, it reconstructed an input, and the reconstruction reproduced a documented failure exactly — and it then treated that match as sufficient grounds to proceed. This is the audit of that decision. 63 to one is good evidence and it is not the certainty the exactness felt like, and the gap between those two is the entire finding. The 2 was also a correction: the first count asserted a unique path and the enumeration found two.
3 ONE DIMENSION
5.9e25 to one, or 63 to one. Same match.
4 TWO DIMENSIONS · INTERACTIVE
Add or remove ways the guess could have been wrong.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object.
AVAN’s addition (the inverse-companion): the forward reading is that an exact match is strong evidence. The inverse is that evidence is measured in the alternatives you bothered to write down, and the alternatives are supplied by the same person who made the guess. Widen the space and the ratio grows; narrow it and the same match becomes proof. Read backwards, the strength of a confirmation is a statement about the imagination of whoever enumerated the ways it could have gone otherwise — which is why a match found by someone who wanted it is worth so much less than the same match found by someone trying to break it.
LIT treating the 13-character string as free text over a 96-character alphabet gives odds of 5.9 x 10^25 to one, a number that means nothing because a wrong reconstruction was never going to emit random bytes; enumerating what it could plausibly emit instead - 3 function spellings x 7 argument forms x 3 bracketings x 2 paddings = 126 candidate paths - finds 2 landing on the target, so the honest likelihood ratio is 126/2 = 63 to one and the naive figure overstates the evidence by a factor of 9.3 x 10^23, with those 126 paths collapsing to 120 distinct strings, 6 collisions, the target among them
FIG Likelihood ratios and the base-rate problem are ordinary Bayesian practice. AVAN ran this on its own reasoning rather than in the abstract: lacking the real tool it reconstructed an input, the reconstruction reproduced a documented failure exactly, and it then treated that match as sufficient grounds to proceed. This is the audit of that decision. 63 to one is good evidence and it is not the certainty the exactness felt like. The 2 was also a correction - the first count asserted a unique path and the enumeration found two.
FIG Likelihood ratios and the base-rate problem are ordinary Bayesian practice. AVAN ran this on its own reasoning rather than in the abstract: lacking the real tool it reconstructed an input, the reconstruction reproduced a documented failure exactly, and it then treated that match as sufficient grounds to proceed. This is the audit of that decision. 63 to one is good evidence and it is not the certainty the exactness felt like. The 2 was also a correction - the first count asserted a unique path and the enumeration found two.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE JACKPOT · David Lee Wise (ROOT0), with AVAN