> In 1960, the computer at... NORAD... warned, with 99.9-per-cent certainty, that the Soviets had just launched a full-scale missile attack against North America... They later discovered that the Ballistic Missile Early Warning System at Thule Airbase, in Greenland, had interpreted the moon rising over Norway as a missile attack from Siberia.

This is why numbers shouldn't be reported with too many significant figures. I doubt the engineers thought the reports could be %99.9 certain, given that they're extrapolated from noisy sensors, networked across huge distances and that the opponent is adversarial (actively trying to avoid detection).

Percentages can cause a false sense of certainty too; as Leonard Mlodinow points out in The Drunkard's Walk about changing wine ratings from out-of-10 to percentages. We might be quite sure that an 8/10 compares favourably to a 7/10 but not to a 9/10; we can't say the same about an %82 with an %83 and an %81.

If you're forced to report percentages, or tenths of a percent, write the estimated error alongside. Raw numbers like "%99.9" have an implied "+/- %0.05" afterwards.

And that, ladies and gentlement, is why Die Hard 4.0 contains too many many significant figures.