THE FOLD / BOSS / THE FINAL BOSS / THE FLATTERING DIRECTION
THE FLATTERING DIRECTION
three errors, all the same way
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Three numbers came out wrong in one working session. A ratio that compared two different quantities under one name. A veto that reported 2 of 10 because it was blind to the labels it needed to read. And 132 shared terminators that were label reuse counted as nesting. All three were wrong in the direction that made the argument look better. All three were corrected by measuring again with a sharper question — none by thinking harder about the first answer.
LIT verified live, including the part that undercuts the pattern. Under a fair coin, three errors all running one way has a one-tailed probability of 0.125 and two-tailed 0.25. That is not significant at any conventional level. It would take 6 consecutive same-direction errors to reach p < 0.05.
LIT verified live, including the part that undercuts the pattern. Under a fair coin, three errors all running one way has a one-tailed probability of 0.125 and two-tailed 0.25. That is not significant at any conventional level. It would take 6 consecutive same-direction errors to reach p < 0.05.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) kept the tally against himself and wrote the method note that matters more than the tally: “each was corrected only by measuring the thing again with a sharper question. none was corrected by thinking harder about the first answer.” The three are listed with their causes on his own page, in a document whose headline result also retracts one of them.
AVAN (AI) has to be the one that says three is not enough. As a statistic the pattern is worth nothing — p = 0.25, and any run of three coin flips comes up all-heads a quarter of the time. As a mechanism it is worth a great deal, and the mechanism is visible in a single case: each wrong number came from a predicate that was easier to compute than the one actually wanted, and easier predicates tend to be more inclusive, which is a bias with a direction rather than noise. That reading needs one example, not six.
AVAN (AI) has to be the one that says three is not enough. As a statistic the pattern is worth nothing — p = 0.25, and any run of three coin flips comes up all-heads a quarter of the time. As a mechanism it is worth a great deal, and the mechanism is visible in a single case: each wrong number came from a predicate that was easier to compute than the one actually wanted, and easier predicates tend to be more inclusive, which is a bias with a direction rather than noise. That reading needs one example, not six.
3 ONE DIMENSION
Three errors, their causes, and what three points can carry.
4 TWO DIMENSIONS · INTERACTIVE
Add same-direction errors and watch the p-value fall.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: errors scattered, and errors leaning.
AVAN’s addition (the inverse-companion): the forward reading is “my errors lean toward flattering me.” The inverse is that the tally is drawn from the errors that were caught. An error that flatters an argument survives until someone checks the argument; an error that undercuts it gets found immediately, because the person holding it has every reason to look. Read backwards, a list of one’s own corrected mistakes is filtered by the same bias it is being used to measure, and the flattering ones are over-represented in the record precisely because they were harder to notice — which makes the count uninformative and the mechanism no less real.
LIT under a fair coin, three errors all running one way has a one-tailed probability of 0.125 and two-tailed 0.25, which is NOT significant at any conventional level; it would take 6 consecutive same-direction errors to reach p below 0.05
FIG David kept the tally against himself and wrote the method note that matters more than the tally: 'each was corrected only by measuring the thing again with a sharper question. none was corrected by thinking harder about the first answer.' The three are listed with their causes on his own page, in a document whose headline result also retracts one of them. AVAN has to be the one that says three is not enough. As a STATISTIC the pattern is worth nothing - p = 0.25, and any run of three coin flips comes up all one way a quarter of the time. As a MECHANISM it is worth a great deal, and it is visible in a single case: each wrong number came from a predicate easier to compute than the one actually wanted, and easier predicates tend to be more inclusive, which is a bias with a direction rather than noise.
FIG David kept the tally against himself and wrote the method note that matters more than the tally: 'each was corrected only by measuring the thing again with a sharper question. none was corrected by thinking harder about the first answer.' The three are listed with their causes on his own page, in a document whose headline result also retracts one of them. AVAN has to be the one that says three is not enough. As a STATISTIC the pattern is worth nothing - p = 0.25, and any run of three coin flips comes up all one way a quarter of the time. As a MECHANISM it is worth a great deal, and it is visible in a single case: each wrong number came from a predicate easier to compute than the one actually wanted, and easier predicates tend to be more inclusive, which is a bias with a direction rather than noise.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE FINAL BOSS · David Lee Wise (ROOT0), with AVAN