Rule

At a high success rate, information is asymmetric: failures teach at many times the rate of
successes. Weight the rare side accordingly - one loss at a 90% win rate deserves roughly nine
wins' worth of attention, and "no problem observed" must be discounted by how few chances a
problem has had to show itself.

When it applies

Judging any high-win-rate strategy (short premium above all); judging whether a change improved
anything; reading a clean track record; interpreting "no difference detected" from any
comparison or test.

What it means

Two consequences, one for trades and one for self-improvement. For trades: a short-premium book
hands you nine confirmations per disconfirmation, so absence of loss is weak evidence of absence
of risk - the book looks safest precisely when the least information has arrived, which is the
classic mechanism of being carried out. For self-improvement: at a high base rate, evidence that
a change HARMS accumulates fast while evidence that it HELPS accumulates slowly, so degradation
should be acted on quickly and "not proven better" should never be read as "not better".

Evidence

Measured on this system's own A/B machinery, 9 paired trials at an ~89% base success rate: a
degrading variant had ~89 percentage points of room to reveal itself and an improving one ~11,
so the arms sat statistically indistinguishable (posterior 0.38 after 45 candidates per side)
despite real differences plausibly existing. The same arithmetic governs a credit spread
collecting $51 against $449 of risk: it needs ~90% accuracy to break even, and its failures
arrive too rarely to teach you before one of them is expensive.