Fischhoff's 1975 method was simple enough to be almost unfair. Give people an obscure historical scenario — his was a British campaign against the Gurkhas of Nepal in 1814 — and ask how likely each possible ending was. Tell some of them which ending actually occurred, and don't tell the others. The informed group consistently rated the real outcome as having been substantially more probable, from the same facts.

The common summary of this is that we misremember our own predictions, which is true but is only the ground floor. Roese and Vohs later separated three stacked levels, and the higher two survive even when your original prediction is sitting in front of you in your own handwriting: it had to happen (inevitability), and it was foreseeable (foreseeability). Those aren't claims about your memory. They're claims about the world, and knowing the ending quietly rewrites them.

The consequence is that hindsight bias is expensive in exactly the places that matter most. A post-mortem conducted after the failure is known will keep concluding that the failure was obvious — which gets squarely in the way of finding out what was genuinely knowable beforehand, unless somebody deliberately works through how it could plausibly have gone the other way. And in a negligence case, jurors who know how it ended are being asked to judge what a person should have foreseen, from a vantage point where foreseeing it is effortless.