THE FOLD / CHEAT / THE EXPLOIT / THE HUNDRED AND THIRTY-TWO
THE HUNDRED AND THIRTY-TWO
label reuse counted as nesting
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A first count found 132 shared loop terminators in BLAS. The test was “a
LIT verified live by building both tests. On sequential reuse the loose test fires 1 and the strict test fires 0; on true nesting both fire 1. Scaled to 310 files with three harmless reuses each, the loose test reports 310 and the strict test reports 0 — and the loose test is a strict superset, so it can never under-count.
DO label appears more than once in a file” — which also matches two sequential loops reusing label 10, legal and harmless and not the trap. Counting the ones where the same label is open twice at once gives 0. The error was not small. It was large, and it ran in the direction that made the argument look good.LIT verified live by building both tests. On sequential reuse the loose test fires 1 and the strict test fires 0; on true nesting both fire 1. Scaled to 310 files with three harmless reuses each, the loose test reports 310 and the strict test reports 0 — and the loose test is a strict superset, so it can never under-count.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) published the retraction with the number attached and named the asymmetry himself: “MY FIRST COUNT SAID 132 SHARED TERMINATORS IN BLAS. it was wrong… the number was not small and wrong. it was large and wrong, in the direction that made my earlier argument look good.”
AVAN (AI) built both predicates to show the failure is structural rather than accidental. “Appears twice in a file” and “open twice at once” are not two attempts at the same question — the first is a property of the text and the second a property of the execution nesting, and one contains the other. Any test that measures the containing set will over-report by exactly the cases that separate them, every time, in the same direction. That is a fact about the predicates, and no amount of care in applying the loose one would have helped.
AVAN (AI) built both predicates to show the failure is structural rather than accidental. “Appears twice in a file” and “open twice at once” are not two attempts at the same question — the first is a property of the text and the second a property of the execution nesting, and one contains the other. Any test that measures the containing set will over-report by exactly the cases that separate them, every time, in the same direction. That is a fact about the predicates, and no amount of care in applying the loose one would have helped.
3 ONE DIMENSION
Two predicates, and the cases that separate them.
4 TWO DIMENSIONS · INTERACTIVE
Switch between sequential reuse and true nesting.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: one set properly containing another.
AVAN’s addition (the inverse-companion): the forward reading is “the loose predicate over-counted.” The inverse is that a superset test is the correct first move and is supposed to over-count. Cheap over-inclusive filters exist precisely so that nothing is missed, and the discipline they require is that their output is treated as a candidate list rather than a result. Read backwards, the failure was not in choosing the loose predicate — it was in reporting its output as the answer, and the same 132 would have been entirely respectable one line earlier, labelled as the set still to be checked.
LIT on sequential reuse the loose test fires 1 and the strict test fires 0, while on true nesting both fire 1; scaled to 310 files with three harmless reuses each the loose test reports 310 and the strict test reports 0; and the loose test is a strict superset, so it can never under-count and its error has a fixed direction
FIG David published the retraction with the number attached and named the asymmetry himself: 'MY FIRST COUNT SAID 132 SHARED TERMINATORS IN BLAS. it was wrong... the number was not small and wrong. it was large and wrong, in the direction that made my earlier argument look good.' AVAN built both predicates to show the failure is structural rather than accidental: 'appears twice in a file' and 'open twice at once' are not two attempts at the same question - the first is a property of the TEXT and the second a property of the EXECUTION NESTING, and one contains the other. Any test that measures the containing set over-reports by exactly the cases that separate them, every time, in the same direction.
FIG David published the retraction with the number attached and named the asymmetry himself: 'MY FIRST COUNT SAID 132 SHARED TERMINATORS IN BLAS. it was wrong... the number was not small and wrong. it was large and wrong, in the direction that made my earlier argument look good.' AVAN built both predicates to show the failure is structural rather than accidental: 'appears twice in a file' and 'open twice at once' are not two attempts at the same question - the first is a property of the TEXT and the second a property of the EXECUTION NESTING, and one contains the other. Any test that measures the containing set over-reports by exactly the cases that separate them, every time, in the same direction.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE EXPLOIT · David Lee Wise (ROOT0), with AVAN