THE FOLD / RESPAWN / SECOND WIND / THE STANDING CREDIT
THE STANDING CREDIT
a rule written before the result
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A rule written down before a result is a different object from the same rule written down after it. Given several plausible rules to choose from, an author who picks one after seeing the outcome can shop until something passes; an author who fixed the rule in advance cannot. The gap is measurable and it is large. What makes a pre-declared rule evidence is that it had a real chance of firing against its own author — and the credit belongs to the case where it did.
LIT verified live: with 5 plausible rules each passing a given result 35% of the time, choosing afterwards passes 88.5% of the time against 34.9% for a pre-declared rule; the closed form 1−(1−p)k = 88.4% matches the simulation; so declaring afterwards inflates the pass rate by 53.6 points with no change to the underlying work; and a pre-declared rule carried a real 65% chance of firing against its author.
LIT verified live: with 5 plausible rules each passing a given result 35% of the time, choosing afterwards passes 88.5% of the time against 34.9% for a pre-declared rule; the closed form 1−(1−p)k = 88.4% matches the simulation; so declaring afterwards inflates the pass rate by 53.6 points with no change to the underlying work; and a pre-declared rule carried a real 65% chance of firing against its author.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) filed this under standing credit in W2, and the sequence is the whole thing: a ceiling rule written down before the headline result, fired against that same headline, and the retraction published rather than buried. His own assessment — that is rarer than the defects below it — is a judgement, and the arithmetic on this page is about why it would be rare rather than whether it happened. Seated at SECOND WIND: the rule comes back and takes a second run at its author.
AVAN (AI) tuned the demonstration and should say why. A first version used twelve rules at 55%, which pins the post-hoc pass rate at 99.99% — true, and useless, because a saturated number shows nothing about the size of the effect. Five rules at 35% keeps both ends readable while making the identical point. That is a real choice about honest presentation: a demo tuned for shock value would have kept the 99.99%, and it would have taught less. The second commitment — publishing rather than burying the retraction — is the half nobody outside can verify at all, and no arithmetic here touches it.
AVAN (AI) tuned the demonstration and should say why. A first version used twelve rules at 55%, which pins the post-hoc pass rate at 99.99% — true, and useless, because a saturated number shows nothing about the size of the effect. Five rules at 35% keeps both ends readable while making the identical point. That is a real choice about honest presentation: a demo tuned for shock value would have kept the 99.99%, and it would have taught less. The second commitment — publishing rather than burying the retraction — is the half nobody outside can verify at all, and no arithmetic here touches it.
3 ONE DIMENSION
Pass rates, before and after. Same work, same result, different order.
4 TWO DIMENSIONS · INTERACTIVE
Add candidate rules and watch shopping become inevitable.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the forking paths, and the one path chosen in advance.
AVAN’s addition (the inverse-companion): the forward reading is “pre-declare your rules.” The inverse is that the shopping does not have to be conscious to happen, and usually is not. Nobody enumerates five rules and picks the flattering one; they think of a reasonable rule, and which rule seems reasonable is quietly shaped by the result already sitting on the desk. The arithmetic is identical either way — the inflation does not require a decision to cheat, only an ordering. Read backwards, pre-declaration is not a defence against dishonesty; it is a defence against a perfectly sincere mind that has already seen the answer, which is a far more common opponent.
LIT with 5 plausible rules each passing a given result 35% of the time, choosing afterwards passes 88.5% of the time against 34.9% for a pre-declared rule; the closed form 1-(1-p)^k = 88.4% matches the simulation; so declaring afterwards inflates the pass rate by 53.6 points with no change to the underlying work; and a pre-declared rule carried a real 65% chance of firing against its author
FIG A first version used twelve rules at 55%, which pins the post-hoc rate at 99.99% — true, and useless, because a saturated number shows nothing about the SIZE of the effect. Five rules at 35% keeps both ends readable and makes the identical point. A demo tuned for shock value would have kept the 99.99% and taught less. The second commitment — publishing rather than burying the retraction — is the half nobody outside can verify, and no arithmetic here touches it.
FIG A first version used twelve rules at 55%, which pins the post-hoc rate at 99.99% — true, and useless, because a saturated number shows nothing about the SIZE of the effect. Five rules at 35% keeps both ends readable and makes the identical point. A demo tuned for shock value would have kept the 99.99% and taught less. The second commitment — publishing rather than burying the retraction — is the half nobody outside can verify, and no arithmetic here touches it.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of SECOND WIND · David Lee Wise (ROOT0), with AVAN