THE FOLD / RESPAWN / THE PHOENIX / THE SELF-CAUGHT SHARE
THE SELF-CAUGHT SHARE
whose control actually fired
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Most of the findings in an audit turned out to have been surfaced by safety checks the subject had already built and left switched on. That number is a compliment, not an indictment — and it bounds what the audit itself contributed. If the subject’s own controls already catch a fraction of defects, an auditor re-running the same checks can only add the ones those controls miss, and that ceiling falls fast as the subject gets better.
LIT verified live: of 8 findings, 5 were surfaced by the subject’s own controls and 3 by the auditor — 62.5% self-caught; so most of what the audit reported was the subject’s instruments working; the ceiling on novel findings is (1−c)×recall, giving 72%, 45%, 18%, 5% at c = 0.2, 0.5, 0.8, 0.95; and a 200,000-defect simulation at c = 0.625 returns a 62.5% self-caught share, matching the arithmetic.
LIT verified live: of 8 findings, 5 were surfaced by the subject’s own controls and 3 by the auditor — 62.5% self-caught; so most of what the audit reported was the subject’s instruments working; the ceiling on novel findings is (1−c)×recall, giving 72%, 45%, 18%, 5% at c = 0.2, 0.5, 0.8, 0.95; and a 200,000-defect simulation at c = 0.625 returns a 62.5% self-caught share, matching the arithmetic.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) built W2 so that it does not flatter the auditor — his phrase — and opened it with the line that reframes the whole exercise: an audit is not one person catching another person. Seated at THE PHOENIX: the subject’s controls were built earlier, left running, and did the work again when nobody was watching.
AVAN (AI) notes the uncomfortable corollary, since a window built not to flatter the auditor should carry it. The better the subject, the less an audit can contribute, and at a self-catch rate of 0.95 an auditor with 90% recall adds 5%. So the audits that produce the most findings are the ones performed on the weakest subjects, and a long list of findings is at least as much a measurement of who was audited as of who did the auditing. The honest way to read a big report is not look what they found but look what was not already being caught — and those are very different sentences.
AVAN (AI) notes the uncomfortable corollary, since a window built not to flatter the auditor should carry it. The better the subject, the less an audit can contribute, and at a self-catch rate of 0.95 an auditor with 90% recall adds 5%. So the audits that produce the most findings are the ones performed on the weakest subjects, and a long list of findings is at least as much a measurement of who was audited as of who did the auditing. The honest way to read a big report is not look what they found but look what was not already being caught — and those are very different sentences.
3 ONE DIMENSION
Eight findings, tagged by whose instrument surfaced them.
4 TWO DIMENSIONS · INTERACTIVE
Improve the subject and watch the audit's ceiling fall.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: two nets over one stream, and the overlap between them.
AVAN’s addition (the inverse-companion): the forward reading is “credit the subject’s own controls.” The inverse is that the finding count is the wrong output entirely. It conflates two quantities that move in opposite directions — how much was wrong, and how much was already being caught — and reports their difference as though it measured the auditor. A report of three findings could mean a careful subject or a lazy audit, and nothing in the number distinguishes them. Read backwards, an audit should publish its self-caught share alongside its findings, because that single ratio is what makes the finding count interpretable at all.
LIT of 8 findings, 5 were surfaced by the subject's own controls and 3 by the auditor — 62.5% self-caught; so most of what the audit reported was the subject's instruments working; the ceiling on novel findings is (1-c) x recall, giving 72%, 45%, 18%, 5% at c = 0.2, 0.5, 0.8, 0.95; and a 200,000-defect simulation at c = 0.625 returns a 62.5% self-caught share, matching
FIG The uncomfortable corollary, carried because the window was built not to flatter the auditor: the better the subject, the LESS an audit can contribute. At a self-catch rate of 0.95 an auditor with 90% recall adds 5%. So audits producing the most findings are performed on the weakest subjects, and a long list measures who was audited at least as much as who audited. The honest reading of a big report is not 'look what they found' but 'look what was not already being caught'.
FIG The uncomfortable corollary, carried because the window was built not to flatter the auditor: the better the subject, the LESS an audit can contribute. At a self-catch rate of 0.95 an auditor with 90% recall adds 5%. So audits producing the most findings are performed on the weakest subjects, and a long list measures who was audited at least as much as who audited. The honest reading of a big report is not 'look what they found' but 'look what was not already being caught'.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE PHOENIX · David Lee Wise (ROOT0), with AVAN