◀ THE FOLD0ROOT.AI // WORLD II · CHEAT · THE BACKDOOR◆ .dlw.fold
THE FOLD / CHEAT / THE BACKDOOR / THE UNVERIFIED SURFACE

THE UNVERIFIED SURFACE

coverage reports on the covered
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A figure a checker cannot reach is not lightly checked. It is unchecked, permanently, and it drifts at whatever rate the work moves — while the coverage number beside it keeps reporting how well everything else is doing.

LIT verified live. take 120 figures of which 118 lie inside the verifier’s reach, each touched with probability 0.02 per edit round over 50 rounds. Coverage reads 98.3%. The probability that at least one of the two unreachable figures has silently drifted is 86.7% — closed form 1−(1−p)(F−V)R, confirmed by simulation at 85.8%. The exposure is exponential in the number uncovered, not in the fraction covered, so 98.3% and 100% are not neighbours: at full coverage the probability is exactly 0.
2 HOW IT WAS WEAVED · AI + HUMAN
The arithmetic is elementary; what it is applied to is not.

AVAN (AI) built this after finding a wrong number on line 5 of a document that shipped beside 73 checks, 47 crosschecks and a 6-of-6 mutant gate. None of them read that file. The model says why that is not bad luck: coverage is reported as a fraction, and a fraction hides the only quantity that matters, which is the raw count of things nothing looks at. Two is a small number and 86.7% is not a small probability.
3 ONE DIMENSION
98.3% covered. 86.7% chance something drifted.
4 TWO DIMENSIONS · INTERACTIVE
Slide the coverage and watch the exposure refuse to fall.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object.
AVAN’s addition (the inverse-companion): the forward reading is that more coverage means less drift. The inverse is that a coverage percentage is a statement about the covered, and the risk lives entirely in the complement. Going from 110 to 118 figures moves coverage from 91.7% to 98.3% and moves the drift probability from 99.996% to 86.7% — a heroic-looking gain that leaves the situation almost unchanged. Read backwards, every coverage metric is computed by the very instrument whose blind spot is the question, and it can only ever report on the part of the world it can see.
LIT take 120 figures of which 118 lie inside the verifier's reach, each touched with probability 0.02 per edit round over 50 rounds: coverage reads 98.3% while the probability that at least one of the two unreachable figures has silently drifted is 86.7% - closed form 1-(1-p)^((F-V)R), confirmed by simulation at 85.8% - because the exposure is exponential in the number UNCOVERED, not in the fraction covered, so 98.3% and 100% are not neighbours and only at full coverage is the probability exactly 0

FIG The arithmetic is elementary; what it is applied to is not. AVAN built this after finding a wrong number on line 5 of a document that shipped beside 73 checks, 47 crosschecks and a 6-of-6 mutant gate, none of which read that file. The model says why that is not bad luck: coverage is reported as a fraction, and a fraction hides the only quantity that matters, which is the raw count of things nothing looks at.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE BACKDOOR · David Lee Wise (ROOT0), with AVAN