We are often more certain than the evidence really allows.
More quotesEntry one, 10am. Forecast: the launch will succeed. Confidence: high.
Nobody records what success means.
Entry two. Repeat use is weak. The forecast becomes “directionally right” because sign-ups were respectable. A statement that sounded perfectly testable has acquired a side exit.
Feynman’s degrees of certainty remove that luxury. Seventy per cent confidence, repeat use above a stated threshold, measured after six weeks: now the claim can meet reality without either party changing its name.
This is less emotionally satisfying than certainty. Organisations often promote the person who says “will” over the person who says “probably”, even when the second person is better calibrated. Firm syntax gets confused with firm evidence. Doubt, however carefully measured, sounds like a character flaw.
The opposite performance is just as empty. “Anything could happen” has no calibration at all. It cannot lose because it never entered. Useful uncertainty has edges: a number, conditions, a time period and an observation that would force revision.
After twenty forecasts, the record becomes interesting. The 70% calls succeeded fourteen times. The 95% forecaster has explained six surprises and remembers none of the original percentages.
Quarter two adds another column: what would have changed the estimate? Some rows contain real conditions—supplier approval, repeat purchase, rainfall. Others say “new information”, which is uncertainty wearing an access badge. The record distinguishes people who expected revision from people who merely wanted permission to revise later.
Another launch enters the sheet. Probability: 60%. Cheap to test, costly to scale, easy to reverse. The six-week test is approved. The outcome column remains blank.