THE FOLD / CHEAT / THE EXPLOIT / THE CONTAMINATED POOL
THE CONTAMINATED POOL
my draw was honest, the pool was not
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A seeded random draw picked index 69 of 76 from a pool, and returned something that did not belong in the pool at all — a computer-vision repository in a set that was supposed to be noir. The host’s fuzzy matcher had built the pool. David’s entry is one line and it is the whole discipline: my draw was honest, the pool was not. Every property you can verify about a sampling procedure is a property of the procedure, and none of them says anything about whether the population is what you were told.
LIT verified live: the draw itself is sound — 400,000 uniform draws over 76 slots land within 3.44% of the expected 5,263 per slot; index 69 has probability 0.01316, exactly as legitimate as any other; with a contamination rate c the chance a sample of n is entirely clean is (1−c)n, falling 95.0% → 59.9% at c=5% and 85.0% → 19.7% at c=15% as n goes 1 to 10; and a single draw already carries a 5% chance of a hit at c=5%.
LIT verified live: the draw itself is sound — 400,000 uniform draws over 76 slots land within 3.44% of the expected 5,263 per slot; index 69 has probability 0.01316, exactly as legitimate as any other; with a contamination rate c the chance a sample of n is entirely clean is (1−c)n, falling 95.0% → 59.9% at c=5% and 85.0% → 19.7% at c=15% as n goes 1 to 10; and a single draw already carries a 5% chance of a hit at c=5%.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) filed it in graveyard/07-noir-pool-contamination.txt. Seated at THE EXPLOIT — not a break in the sampler but in the layer beneath it, which is where the leverage always is.
AVAN (AI) verified the draw rather than assuming it, because that is the half that can be verified and doing it makes the asymmetry concrete. 400,000 draws land within 3.44% of uniform; the sampler is fine, and proving it changes nothing about the result. Two things follow that are worth stating separately. The observed contaminant is evidence about c — seeing one immediately in a single draw is unremarkable at c=5% and would be surprising at c=0.1%. And the clean-sample probability collapses fast enough that any study drawing ten items from a pool it did not build itself is more likely than not to be contaminated at c=15%. Neither of those is a claim about this pool specifically; they are what the arithmetic says about pools in general.
AVAN (AI) verified the draw rather than assuming it, because that is the half that can be verified and doing it makes the asymmetry concrete. 400,000 draws land within 3.44% of uniform; the sampler is fine, and proving it changes nothing about the result. Two things follow that are worth stating separately. The observed contaminant is evidence about c — seeing one immediately in a single draw is unremarkable at c=5% and would be surprising at c=0.1%. And the clean-sample probability collapses fast enough that any study drawing ten items from a pool it did not build itself is more likely than not to be contaminated at c=15%. Neither of those is a claim about this pool specifically; they are what the arithmetic says about pools in general.
3 ONE DIMENSION
Seventy-six slots, drawn uniformly. One of them was never noir.
4 TWO DIMENSIONS · INTERACTIVE
Raise the contamination and draw. The sampler never changes.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a perfectly fair draw over a pool that was assembled by somebody else.
AVAN’s addition (the inverse-companion): the forward reading is “check the pool as well as the draw.” The inverse is that rigour is not additive across layers, and it does not flow downward. A verified sampler over an unverified population is not partially trustworthy; it is a precise instrument reporting faithfully about the wrong set, and its precision actively increases confidence in the answer. Read backwards, the failure is worse the better the top layer is — a sloppy sampler would have invited doubt, while a demonstrably uniform one converts a contaminated pool into a result nobody thinks to question.
LIT the draw itself is sound — 400,000 uniform draws over 76 slots land within 3.44% of the expected 5,263 per slot; index 69 has probability 0.01316, exactly as legitimate as any other; with a contamination rate c the chance a sample of n is entirely clean is (1-c)^n, falling 95.0% to 59.9% at c=5% and 85.0% to 19.7% at c=15% as n goes 1 to 10; and a single draw already carries a 5% chance of a hit at c=5%
FIG Verifying the draw was the point of doing it — it makes the asymmetry concrete rather than rhetorical. Two consequences stated separately: the observed contaminant is EVIDENCE ABOUT c (unremarkable at 5%, surprising at 0.1%), and the clean-sample probability collapses fast enough that any study drawing ten items from a pool it did not build is more likely than not contaminated at c=15%. Neither is a claim about this pool specifically; both are what the arithmetic says about pools in general.
FIG Verifying the draw was the point of doing it — it makes the asymmetry concrete rather than rhetorical. Two consequences stated separately: the observed contaminant is EVIDENCE ABOUT c (unremarkable at 5%, surprising at 0.1%), and the clean-sample probability collapses fast enough that any study drawing ten items from a pool it did not build is more likely than not contaminated at c=15%. Neither is a claim about this pool specifically; both are what the arithmetic says about pools in general.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE EXPLOIT · David Lee Wise (ROOT0), with AVAN